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

Comparing libeio/eio.c (file contents):
Revision 1.3 by root, Sun May 11 00:06:25 2008 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 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#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
… …
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <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
14#include <dirent.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);
15 81
16#ifndef EIO_FINISH 82#ifndef EIO_FINISH
17# 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
18#endif 84#endif
19 85
… …
23 89
24#ifndef EIO_FEED 90#ifndef EIO_FEED
25# 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)
26#endif 92#endif
27 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
28#ifdef _WIN32 105#ifdef _WIN32
29 106
30 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
31 196
32#else 197#else
33 198
34# include "config.h"
35# include <sys/time.h> 199 #include <sys/time.h>
36# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
37# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
38# include <utime.h> 237# include <utime.h>
39# include <signal.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 238#endif
44 239
45#endif 240#if HAVE_SYS_SYSCALL_H
46 241# include <sys/syscall.h>
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif 242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
246#endif
50 247
51#if HAVE_SENDFILE 248#if HAVE_SENDFILE
52# if __linux 249# if __linux
53# include <sys/sendfile.h> 250# include <sys/sendfile.h>
54# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 252# include <sys/socket.h>
56# include <sys/uio.h> 253# include <sys/uio.h>
57# elif __hpux 254# elif __hpux
58# include <sys/socket.h> 255# include <sys/socket.h>
59# elif __solaris /* not yet */ 256# elif __solaris
60# include <sys/sendfile.h> 257# include <sys/sendfile.h>
61# else 258# else
62# error sendfile support requested but not available 259# error sendfile support requested but not available
63# endif 260# endif
64#endif 261#endif
65 262
66/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
68 272
69/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 274#ifndef NAME_MAX
71# define NAME_MAX 4096 275# define NAME_MAX 4096
72#endif 276#endif
73 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
74/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
76 285
77#define dBUF \ 286#define dBUF \
78 char *eio_buf; \
79 X_LOCK (wrklock); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 288 errno = ENOMEM; \
82 if (!eio_buf) \ 289 if (!eio_buf) \
83 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
84 294
85#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
340
341#define ETP_PRI_MIN EIO_PRI_MIN
342#define ETP_PRI_MAX EIO_PRI_MAX
343
344struct etp_worker;
345
346#define ETP_REQ eio_req
347#define ETP_DESTROY(req) eio_destroy (req)
348static int eio_finish (eio_req *req);
349#define ETP_FINISH(req) eio_finish (req)
350static void eio_execute (struct etp_worker *self, eio_req *req);
351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
352
353/*****************************************************************************/
354
355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
356
357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
359tvdiff (struct timeval *tv1, struct timeval *tv2)
360{
361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
363}
364
365static unsigned int started, idle, wanted = 4;
86 366
87static void (*want_poll_cb) (void); 367static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
89 369
90static unsigned int max_poll_time = 0; 370static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 371static unsigned int max_poll_reqs; /* reslock */
92 372
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 373static unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 374static unsigned int nready; /* reqlock */
95{ 375static unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 378
100static unsigned int started, idle, wanted; 379static xmutex_t wrklock;
380static xmutex_t reslock;
381static xmutex_t reqlock;
382static xcond_t reqwait;
383
384#if !HAVE_PREADWRITE
385/*
386 * make our pread/pwrite emulation safe against themselves, but not against
387 * normal read/write by using a mutex. slows down execution a lot,
388 * but that's your problem, not mine.
389 */
390static xmutex_t preadwritelock;
391#endif
392
393typedef struct etp_worker
394{
395 struct tmpbuf tmpbuf;
396
397 /* locked by wrklock */
398 struct etp_worker *prev, *next;
399
400 xthread_t tid;
401
402#ifdef ETP_WORKER_COMMON
403 ETP_WORKER_COMMON
404#endif
405} etp_worker;
406
407static etp_worker wrk_first; /* NOT etp */
408
409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 411
102/* worker threads management */ 412/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 413
105typedef struct worker { 414static void
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 415etp_worker_clear (etp_worker *wrk)
120{ 416{
121 if (wrk->dirp) 417}
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126 418
127 if (wrk->dbuf) 419static void ecb_cold
128 { 420etp_worker_free (etp_worker *wrk)
421{
129 free (wrk->dbuf); 422 free (wrk->tmpbuf.ptr);
130 wrk->dbuf = 0;
131 }
132}
133 423
134static void worker_free (worker *wrk)
135{
136 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
138 426
139 free (wrk); 427 free (wrk);
140} 428}
141 429
142static volatile unsigned int nreqs, nready, npending; 430static unsigned int
143static volatile unsigned int max_idle = 4; 431etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 432{
433 int retval;
434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
435 retval = nreqs;
436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 437 return retval;
152} 438}
153 439
154unsigned int eio_nready (void) 440static unsigned int
441etp_nready (void)
155{ 442{
156 unsigned int retval; 443 unsigned int retval;
157 444
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 446 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 448
162 return retval; 449 return retval;
163} 450}
164 451
165unsigned int eio_npending (void) 452static unsigned int
453etp_npending (void)
166{ 454{
167 unsigned int retval; 455 unsigned int retval;
168 456
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 458 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 460
173 return retval; 461 return retval;
174} 462}
175 463
176unsigned int eio_nthreads (void) 464static unsigned int
465etp_nthreads (void)
177{ 466{
178 unsigned int retval; 467 unsigned int retval;
179 468
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 470 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 477 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 478 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 479 * per shift, the most expensive operation.
191 */ 480 */
192typedef struct { 481typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 482 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 483 int size;
195} reqq; 484} etp_reqq;
196 485
197static reqq req_queue; 486static etp_reqq req_queue;
198static reqq res_queue; 487static etp_reqq res_queue;
199 488
200static int reqq_push (reqq *q, eio_req *req) 489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
501reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 502{
202 int pri = req->pri; 503 int pri = req->pri;
203 req->next = 0; 504 req->next = 0;
204 505
205 if (q->qe[pri]) 506 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
212 513
213 return q->size++; 514 return q->size++;
214} 515}
215 516
216static eio_req *reqq_shift (reqq *q) 517static ETP_REQ * ecb_noinline
518reqq_shift (etp_reqq *q)
217{ 519{
218 int pri; 520 int pri;
219 521
220 if (!q->size) 522 if (!q->size)
221 return 0; 523 return 0;
222 524
223 --q->size; 525 --q->size;
224 526
225 for (pri = EIO_NUM_PRI; pri--; ) 527 for (pri = ETP_NUM_PRI; pri--; )
226 { 528 {
227 eio_req *req = q->qs[pri]; 529 eio_req *req = q->qs[pri];
228 530
229 if (req) 531 if (req)
230 { 532 {
… …
236 } 538 }
237 539
238 abort (); 540 abort ();
239} 541}
240 542
241static void grp_try_feed (eio_req *grp) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
242{ 545{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 546 X_MUTEX_CREATE (wrklock);
244 { 547 X_MUTEX_CREATE (reslock);
245 int old_len = grp->size; 548 X_MUTEX_CREATE (reqlock);
549 X_COND_CREATE (reqwait);
246 550
247 EIO_FEED (grp); 551 reqq_init (&req_queue);
552 reqq_init (&res_queue);
248 553
249 /* stop if no progress has been made */ 554 wrk_first.next =
250 if (old_len == grp->size) 555 wrk_first.prev = &wrk_first;
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256}
257 556
258static int eio_finish (eio_req *req); 557 started = 0;
558 idle = 0;
559 nreqs = 0;
560 nready = 0;
561 npending = 0;
259 562
260static int grp_dec (eio_req *grp) 563 want_poll_cb = want_poll;
261{ 564 done_poll_cb = done_poll;
262 --grp->size;
263 565
264 /* call feeder, if applicable */
265 grp_try_feed (grp);
266
267 /* finish, if done */
268 if (!grp->size && grp->int1)
269 return eio_finish (grp);
270 else
271 return 0; 566 return 0;
272} 567}
273 568
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR2_FREE)
277 free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 req->flags |= EIO_FLAG_CANCELLED;
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 569X_THREAD_PROC (etp_proc);
323 570
324static void start_thread (void) 571static void ecb_cold
572etp_start_thread (void)
325{ 573{
326 worker *wrk = calloc (1, sizeof (worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 575
328 if (!wrk) 576 /*TODO*/
329 croak ("unable to allocate worker thread data"); 577 assert (("unable to allocate worker thread data", wrk));
330 578
331 X_LOCK (wrklock); 579 X_LOCK (wrklock);
332 580
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 581 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 582 {
335 wrk->prev = &wrk_first; 583 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 584 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 585 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 586 wrk_first.next = wrk;
… …
342 free (wrk); 590 free (wrk);
343 591
344 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
345} 593}
346 594
595static void
347static void maybe_start_thread (void) 596etp_maybe_start_thread (void)
348{ 597{
349 if (eio_nthreads () >= wanted) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 599 return;
351 600
352 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 602 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 603 return;
355 604
356 start_thread (); 605 etp_start_thread ();
357} 606}
358 607
359void eio_submit (eio_req *req) 608static void ecb_cold
609etp_end_thread (void)
360{ 610{
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void)
373{
374 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
375 612
376 req->type = EIO_QUIT; 613 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 615
379 X_LOCK (reqlock); 616 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
… …
384 X_LOCK (wrklock); 621 X_LOCK (wrklock);
385 --started; 622 --started;
386 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
387} 624}
388 625
389void eio_set_max_poll_time (double nseconds) 626static int
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 627etp_poll (void)
426{ 628{
427 int maxreqs = max_poll_reqs; 629 unsigned int maxreqs;
630 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
429 eio_req *req;
430 632
633 X_LOCK (reslock);
634 maxreqs = max_poll_reqs;
635 maxtime = max_poll_time;
636 X_UNLOCK (reslock);
637
431 if (max_poll_time) 638 if (maxtime)
432 gettimeofday (&tv_start, 0); 639 gettimeofday (&tv_start, 0);
433 640
434 for (;;) 641 for (;;)
435 { 642 {
643 ETP_REQ *req;
644
436 maybe_start_thread (); 645 etp_maybe_start_thread ();
437 646
438 X_LOCK (reslock); 647 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 648 req = reqq_shift (&res_queue);
440 649
441 if (req) 650 if (req)
… …
449 X_UNLOCK (reslock); 658 X_UNLOCK (reslock);
450 659
451 if (!req) 660 if (!req)
452 return 0; 661 return 0;
453 662
663 X_LOCK (reqlock);
454 --nreqs; 664 --nreqs;
665 X_UNLOCK (reqlock);
455 666
456 if (req->type == EIO_GROUP && req->size) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457 { 668 {
458 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
459 continue; 670 continue;
460 } 671 }
461 else 672 else
462 { 673 {
463 int res = eio_finish (req); 674 int res = ETP_FINISH (req);
464 if (res) 675 if (ecb_expect_false (res))
465 return res; 676 return res;
466 } 677 }
467 678
468 if (maxreqs && !--maxreqs) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
469 break; 680 break;
470 681
471 if (max_poll_time) 682 if (maxtime)
472 { 683 {
473 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
474 685
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 686 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 687 break;
477 } 688 }
478 } 689 }
479 690
480 errno = EAGAIN; 691 errno = EAGAIN;
481 return -1; 692 return -1;
482} 693}
483 694
695static void
696etp_cancel (ETP_REQ *req)
697{
698 req->cancelled = 1;
699
700 eio_grp_cancel (req);
701}
702
703static void
704etp_submit (ETP_REQ *req)
705{
706 req->pri -= ETP_PRI_MIN;
707
708 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
709 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
710
711 if (ecb_expect_false (req->type == EIO_GROUP))
712 {
713 /* I hope this is worth it :/ */
714 X_LOCK (reqlock);
715 ++nreqs;
716 X_UNLOCK (reqlock);
717
718 X_LOCK (reslock);
719
720 ++npending;
721
722 if (!reqq_push (&res_queue, req) && want_poll_cb)
723 want_poll_cb ();
724
725 X_UNLOCK (reslock);
726 }
727 else
728 {
729 X_LOCK (reqlock);
730 ++nreqs;
731 ++nready;
732 reqq_push (&req_queue, req);
733 X_COND_SIGNAL (reqwait);
734 X_UNLOCK (reqlock);
735
736 etp_maybe_start_thread ();
737 }
738}
739
740static void ecb_cold
741etp_set_max_poll_time (double nseconds)
742{
743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
744 max_poll_time = nseconds * EIO_TICKS;
745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
746}
747
748static void ecb_cold
749etp_set_max_poll_reqs (unsigned int maxreqs)
750{
751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
752 max_poll_reqs = maxreqs;
753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
754}
755
756static void ecb_cold
757etp_set_max_idle (unsigned int nthreads)
758{
759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
760 max_idle = nthreads;
761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
762}
763
764static void ecb_cold
765etp_set_idle_timeout (unsigned int seconds)
766{
767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
768 idle_timeout = seconds;
769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
770}
771
772static void ecb_cold
773etp_set_min_parallel (unsigned int nthreads)
774{
775 if (wanted < nthreads)
776 wanted = nthreads;
777}
778
779static void ecb_cold
780etp_set_max_parallel (unsigned int nthreads)
781{
782 if (wanted > nthreads)
783 wanted = nthreads;
784
785 while (started > wanted)
786 etp_end_thread ();
787}
788
789/*****************************************************************************/
790
791static void
792grp_try_feed (eio_req *grp)
793{
794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
795 {
796 grp->flags &= ~EIO_FLAG_GROUPADD;
797
798 EIO_FEED (grp);
799
800 /* stop if no progress has been made */
801 if (!(grp->flags & EIO_FLAG_GROUPADD))
802 {
803 grp->feed = 0;
804 break;
805 }
806 }
807}
808
809static int
810grp_dec (eio_req *grp)
811{
812 --grp->size;
813
814 /* call feeder, if applicable */
815 grp_try_feed (grp);
816
817 /* finish, if done */
818 if (!grp->size && grp->int1)
819 return eio_finish (grp);
820 else
821 return 0;
822}
823
824static void
825eio_destroy (eio_req *req)
826{
827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
829
830 EIO_DESTROY (req);
831}
832
833static int
834eio_finish (eio_req *req)
835{
836 int res = EIO_FINISH (req);
837
838 if (req->grp)
839 {
840 int res2;
841 eio_req *grp = req->grp;
842
843 /* unlink request */
844 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
845 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
846
847 if (grp->grp_first == req)
848 grp->grp_first = req->grp_next;
849
850 res2 = grp_dec (grp);
851
852 if (!res)
853 res = res2;
854 }
855
856 eio_destroy (req);
857
858 return res;
859}
860
861void
862eio_grp_cancel (eio_req *grp)
863{
864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
865 eio_cancel (grp);
866}
867
868void
869eio_cancel (eio_req *req)
870{
871 etp_cancel (req);
872}
873
874void
875eio_submit (eio_req *req)
876{
877 etp_submit (req);
878}
879
880unsigned int
881eio_nreqs (void)
882{
883 return etp_nreqs ();
884}
885
886unsigned int
887eio_nready (void)
888{
889 return etp_nready ();
890}
891
892unsigned int
893eio_npending (void)
894{
895 return etp_npending ();
896}
897
898unsigned int ecb_cold
899eio_nthreads (void)
900{
901 return etp_nthreads ();
902}
903
904void ecb_cold
905eio_set_max_poll_time (double nseconds)
906{
907 etp_set_max_poll_time (nseconds);
908}
909
910void ecb_cold
911eio_set_max_poll_reqs (unsigned int maxreqs)
912{
913 etp_set_max_poll_reqs (maxreqs);
914}
915
916void ecb_cold
917eio_set_max_idle (unsigned int nthreads)
918{
919 etp_set_max_idle (nthreads);
920}
921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
929eio_set_min_parallel (unsigned int nthreads)
930{
931 etp_set_min_parallel (nthreads);
932}
933
934void ecb_cold
935eio_set_max_parallel (unsigned int nthreads)
936{
937 etp_set_max_parallel (nthreads);
938}
939
940int eio_poll (void)
941{
942 return etp_poll ();
943}
944
484/*****************************************************************************/ 945/*****************************************************************************/
485/* work around various missing functions */ 946/* work around various missing functions */
486 947
487#if !HAVE_PREADWRITE 948#if !HAVE_PREADWRITE
949# undef pread
950# undef pwrite
488# define pread aio_pread 951# define pread eio__pread
489# define pwrite aio_pwrite 952# define pwrite eio__pwrite
490 953
491/* 954static eio_ssize_t
492 * make our pread/pwrite safe against themselves, but not against 955eio__pread (int fd, void *buf, size_t count, off_t offset)
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499{ 956{
500 ssize_t res; 957 eio_ssize_t res;
501 off_t ooffset; 958 off_t ooffset;
502 959
503 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
504 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
505 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
508 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
509 966
510 return res; 967 return res;
511} 968}
512 969
513static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 970static eio_ssize_t
971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514{ 972{
515 ssize_t res; 973 eio_ssize_t res;
516 off_t ooffset; 974 off_t ooffset;
517 975
518 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
519 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
520 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
521 res = write (fd, buf, count); 979 res = write (fd, buf, count);
522 lseek (fd, offset, SEEK_SET); 980 lseek (fd, ooffset, SEEK_SET);
523 X_UNLOCK (preadwritelock); 981 X_UNLOCK (preadwritelock);
524 982
525 return res; 983 return res;
526} 984}
527#endif 985#endif
528 986
529#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
530 988
989# undef utimes
531# define utimes(path,times) aio_utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
532# define futimes(fd,times) aio_futimes (fd, times)
533 991
992static int
534static int aio_utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
535{ 994{
536 if (times) 995 if (times)
537 { 996 {
538 struct utimbuf buf; 997 struct utimbuf buf;
539 998
… …
544 } 1003 }
545 else 1004 else
546 return utime (filename, 0); 1005 return utime (filename, 0);
547} 1006}
548 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
549static int aio_futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
550{ 1017{
551 errno = ENOSYS; 1018 errno = ENOSYS;
552 return -1; 1019 return -1;
553} 1020}
554 1021
555#endif 1022#endif
556 1023
557#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
1025# undef fdatasync
558# define fdatasync fsync 1026# define fdatasync(fd) fsync (fd)
559#endif 1027#endif
1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = EIO_ENOSYS ();
1038#endif
1039
1040 if (res < 0 && errno == ENOSYS && fd >= 0)
1041 sync ();
1042
1043 return res;
1044}
1045
1046/* sync_file_range always needs emulation */
1047static int
1048eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1049{
1050#if HAVE_SYNC_FILE_RANGE
1051 int res;
1052
1053 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1054 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1055 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1056 {
1057 flags = 0
1058 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1059 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1060 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1061 }
1062
1063 res = sync_file_range (fd, offset, nbytes, flags);
1064
1065 if (!res || errno != ENOSYS)
1066 return res;
1067#endif
1068
1069 /* even though we could play tricks with the flags, it's better to always
1070 * call fdatasync, as that matches the expectation of its users best */
1071 return fdatasync (fd);
1072}
1073
1074static int
1075eio__fallocate (int fd, int mode, off_t offset, size_t len)
1076{
1077#if HAVE_LINUX_FALLOCATE
1078 return fallocate (fd, mode, offset, len);
1079#else
1080 return EIO_ENOSYS ();
1081#endif
1082}
560 1083
561#if !HAVE_READAHEAD 1084#if !HAVE_READAHEAD
1085# undef readahead
562# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1086# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563 1087
1088static eio_ssize_t
564static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1089eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565{ 1090{
566 size_t todo = count; 1091 size_t todo = count;
567 dBUF; 1092 dBUF;
568 1093
569 while (todo > 0) 1094 while (todo > 0)
… …
573 pread (fd, eio_buf, len, offset); 1098 pread (fd, eio_buf, len, offset);
574 offset += len; 1099 offset += len;
575 todo -= len; 1100 todo -= len;
576 } 1101 }
577 1102
578 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
579 return count; 1107 return 0;
580} 1108}
581 1109
582#endif
583
584#if !HAVE_READDIR_R
585# define readdir_r aio_readdir_r
586
587static mutex_t readdirlock = X_MUTEX_INIT;
588
589static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590{
591 X_DIRENT *e;
592 int errorno;
593
594 X_LOCK (readdirlock);
595
596 e = readdir (dirp);
597 errorno = errno;
598
599 if (e)
600 {
601 *res = ent;
602 strcpy (ent->d_name, e->d_name);
603 }
604 else
605 *res = 0;
606
607 X_UNLOCK (readdirlock);
608
609 errno = errorno;
610 return e ? 0 : -1;
611}
612#endif 1110#endif
613 1111
614/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
1113static eio_ssize_t
615static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
616{ 1115{
1116 eio_ssize_t written = 0;
617 ssize_t res; 1117 eio_ssize_t res;
618 1118
619 if (!count) 1119 if (!count)
620 return 0; 1120 return 0;
621 1121
1122 for (;;)
1123 {
1124#ifdef __APPLE__
1125# undef HAVE_SENDFILE /* broken, as everything on os x */
1126#endif
622#if HAVE_SENDFILE 1127#if HAVE_SENDFILE
623# if __linux 1128# if __linux
1129 off_t soffset = offset;
624 res = sendfile (ofd, ifd, &offset, count); 1130 res = sendfile (ofd, ifd, &soffset, count);
625 1131
626# elif __freebsd 1132# elif __FreeBSD__
627 /* 1133 /*
628 * Of course, the freebsd sendfile is a dire hack with no thoughts 1134 * Of course, the freebsd sendfile is a dire hack with no thoughts
629 * wasted on making it similar to other I/O functions. 1135 * wasted on making it similar to other I/O functions.
630 */ 1136 */
631 {
632 off_t sbytes; 1137 off_t sbytes;
633 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1138 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634 1139
635 if (res < 0 && sbytes) 1140 #if 0 /* according to the manpage, this is correct, but broken behaviour */
636 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1141 /* freebsd' sendfile will return 0 on success */
1142 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1143 /* not on e.g. EIO or EPIPE - sounds broken */
1144 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637 res = sbytes; 1145 res = sbytes;
638 } 1146 #endif
1147
1148 /* according to source inspection, this is correct, and useful behaviour */
1149 if (sbytes)
1150 res = sbytes;
1151
1152# elif defined __APPLE__
1153 off_t sbytes = count;
1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1155
1156 /* according to the manpage, sbytes is always valid */
1157 if (sbytes)
1158 res = sbytes;
639 1159
640# elif __hpux 1160# elif __hpux
641 res = sendfile (ofd, ifd, offset, count, 0, 0); 1161 res = sendfile (ofd, ifd, offset, count, 0, 0);
642 1162
643# elif __solaris 1163# elif __solaris
644 {
645 struct sendfilevec vec; 1164 struct sendfilevec vec;
646 size_t sbytes; 1165 size_t sbytes;
647 1166
648 vec.sfv_fd = ifd; 1167 vec.sfv_fd = ifd;
649 vec.sfv_flag = 0; 1168 vec.sfv_flag = 0;
650 vec.sfv_off = offset; 1169 vec.sfv_off = offset;
651 vec.sfv_len = count; 1170 vec.sfv_len = count;
652 1171
653 res = sendfilev (ofd, &vec, 1, &sbytes); 1172 res = sendfilev (ofd, &vec, 1, &sbytes);
654 1173
655 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
656 res = sbytes; 1175 res = sbytes;
657 }
658 1176
659# endif 1177# endif
1178
1179#elif defined (_WIN32) && 0
1180 /* does not work, just for documentation of what would need to be done */
1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1182 /* libeio guarantees that the file offset does not change, and windows */
1183 /* has no way to get an independent handle to the same file description */
1184 HANDLE h = TO_SOCKET (ifd);
1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1187
660#else 1188#else
661 res = -1; 1189 res = EIO_ENOSYS ();
662 errno = ENOSYS;
663#endif 1190#endif
664 1191
1192 /* we assume sendfile can copy at least 128mb in one go */
1193 if (res <= 128 * 1024 * 1024)
1194 {
1195 if (res > 0)
1196 written += res;
1197
1198 if (written)
1199 return written;
1200
1201 break;
1202 }
1203 else
1204 {
1205 /* if we requested more, then probably the kernel was lazy */
1206 written += res;
1207 offset += res;
1208 count -= res;
1209
1210 if (!count)
1211 return written;
1212 }
1213 }
1214
665 if (res < 0 1215 if (res < 0
666 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1216 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1217 /* BSDs */
1218#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1219 || errno == ENOTSUP
1220#endif
1221#ifdef EOPNOTSUPP /* windows */
1222 || errno == EOPNOTSUPP /* BSDs */
1223#endif
667#if __solaris 1224#if __solaris
668 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1225 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
669#endif 1226#endif
670 ) 1227 )
671 ) 1228 )
… …
675 1232
676 res = 0; 1233 res = 0;
677 1234
678 while (count) 1235 while (count)
679 { 1236 {
680 ssize_t cnt; 1237 eio_ssize_t cnt;
681 1238
682 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1239 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
683 1240
684 if (cnt <= 0) 1241 if (cnt <= 0)
685 { 1242 {
… …
697 1254
698 offset += cnt; 1255 offset += cnt;
699 res += cnt; 1256 res += cnt;
700 count -= cnt; 1257 count -= cnt;
701 } 1258 }
1259
1260 FUBd;
702 } 1261 }
703 1262
704 return res; 1263 return res;
705} 1264}
706 1265
707/* read a full directory */ 1266#ifdef PAGESIZE
708static void scandir_ (eio_req *req, worker *self) 1267# define eio_pagesize() PAGESIZE
1268#else
1269static intptr_t
1270eio_pagesize (void)
709{ 1271{
710 DIR *dirp; 1272 static intptr_t page;
711 union
712 {
713 EIO_STRUCT_DIRENT d;
714 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715 } *u;
716 EIO_STRUCT_DIRENT *entp;
717 char *name, *names;
718 int memlen = 4096;
719 int memofs = 0;
720 int res = 0;
721 1273
722 X_LOCK (wrklock); 1274 if (!page)
723 self->dirp = dirp = opendir (req->ptr1); 1275 page = sysconf (_SC_PAGESIZE);
724 self->dbuf = u = malloc (sizeof (*u));
725 req->flags |= EIO_FLAG_PTR2_FREE;
726 req->ptr2 = names = malloc (memlen);
727 X_UNLOCK (wrklock);
728 1276
729 if (dirp && u && names) 1277 return page;
730 for (;;) 1278}
1279#endif
1280
1281static void
1282eio_page_align (void **addr, size_t *length)
1283{
1284 intptr_t mask = eio_pagesize () - 1;
1285
1286 /* round down addr */
1287 intptr_t adj = mask & (intptr_t)*addr;
1288
1289 *addr = (void *)((intptr_t)*addr - adj);
1290 *length += adj;
1291
1292 /* round up length */
1293 *length = (*length + mask) & ~mask;
1294}
1295
1296#if !_POSIX_MEMLOCK
1297# define eio__mlockall(a) EIO_ENOSYS ()
1298#else
1299
1300static int
1301eio__mlockall (int flags)
1302{
1303 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1304 extern int mallopt (int, int);
1305 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1306 #endif
1307
1308 if (EIO_MCL_CURRENT != MCL_CURRENT
1309 || EIO_MCL_FUTURE != MCL_FUTURE)
1310 {
1311 flags = 0
1312 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1313 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1314 }
1315
1316 return mlockall (flags);
1317}
1318#endif
1319
1320#if !_POSIX_MEMLOCK_RANGE
1321# define eio__mlock(a,b) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__mlock (void *addr, size_t length)
1326{
1327 eio_page_align (&addr, &length);
1328
1329 return mlock (addr, length);
1330}
1331
1332#endif
1333
1334#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1335# define eio__msync(a,b,c) EIO_ENOSYS ()
1336#else
1337
1338static int
1339eio__msync (void *mem, size_t len, int flags)
1340{
1341 eio_page_align (&mem, &len);
1342
1343 if (EIO_MS_ASYNC != MS_SYNC
1344 || EIO_MS_INVALIDATE != MS_INVALIDATE
1345 || EIO_MS_SYNC != MS_SYNC)
1346 {
1347 flags = 0
1348 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1349 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1350 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1351 }
1352
1353 return msync (mem, len, flags);
1354}
1355
1356#endif
1357
1358static int
1359eio__mtouch (eio_req *req)
1360{
1361 void *mem = req->ptr2;
1362 size_t len = req->size;
1363 int flags = req->int1;
1364
1365 eio_page_align (&mem, &len);
1366
1367 {
1368 intptr_t addr = (intptr_t)mem;
1369 intptr_t end = addr + len;
1370 intptr_t page = eio_pagesize ();
1371
1372 if (addr < end)
1373 if (flags & EIO_MT_MODIFY) /* modify */
1374 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1375 else
1376 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 }
1378
1379 return 0;
1380}
1381
1382/*****************************************************************************/
1383/* requests implemented outside eio_execute, because they are so large */
1384
1385static void
1386eio__lseek (eio_req *req)
1387{
1388 /* this usually gets optimised away completely, or your compiler sucks, */
1389 /* or the whence constants really are not 0, 1, 2 */
1390 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1391 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1392 : req->int2 == EIO_SEEK_END ? SEEK_END
1393 : req->int2;
1394
1395 req->offs = lseek (req->int1, req->offs, whence);
1396 req->result = req->offs == (off_t)-1 ? -1 : 0;
1397}
1398
1399/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1400static int
1401eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1402{
1403 const char *rel = path;
1404 char *res;
1405 char *tmp1, *tmp2;
1406#if SYMLOOP_MAX > 32
1407 int symlinks = SYMLOOP_MAX;
1408#else
1409 int symlinks = 32;
1410#endif
1411
1412 errno = EINVAL;
1413 if (!rel)
1414 return -1;
1415
1416 errno = ENOENT;
1417 if (!*rel)
1418 return -1;
1419
1420 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1421 tmp1 = res + PATH_MAX;
1422 tmp2 = tmp1 + PATH_MAX;
1423
1424#if 0 /* disabled, the musl way to do things is just too racy */
1425#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1426 /* on linux we may be able to ask the kernel */
1427 {
1428 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1429
1430 if (fd >= 0)
731 { 1431 {
1432 sprintf (tmp1, "/proc/self/fd/%d", fd);
1433 req->result = readlink (tmp1, res, PATH_MAX);
1434 close (fd);
1435
1436 /* here we should probably stat the open file and the disk file, to make sure they still match */
1437
1438 if (req->result > 0)
1439 goto done;
1440 }
1441 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1442 return;
1443 }
1444#endif
1445#endif
1446
1447 if (*rel != '/')
1448 {
1449 int len;
1450
732 errno = 0; 1451 errno = ENOENT;
733 readdir_r (dirp, &u->d, &entp); 1452 if (wd == EIO_INVALID_WD)
1453 return -1;
1454
1455 if (wd == EIO_CWD)
1456 {
1457 if (!getcwd (res, PATH_MAX))
1458 return -1;
734 1459
735 if (!entp) 1460 len = strlen (res);
1461 }
1462 else
1463 memcpy (res, wd->str, len = wd->len);
1464
1465 if (res [1]) /* only use if not / */
1466 res += len;
1467 }
1468
1469 while (*rel)
1470 {
1471 eio_ssize_t len, linklen;
1472 const char *beg = rel;
1473
1474 while (*rel && *rel != '/')
1475 ++rel;
1476
1477 len = rel - beg;
1478
1479 if (!len) /* skip slashes */
1480 {
736 break; 1481 ++rel;
1482 continue;
1483 }
737 1484
738 name = entp->d_name; 1485 if (beg [0] == '.')
1486 {
1487 if (len == 1)
1488 continue; /* . - nop */
739 1489
740 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1490 if (beg [1] == '.' && len == 2)
1491 {
1492 /* .. - back up one component, if possible */
1493
1494 while (res != tmpbuf->ptr)
1495 if (*--res == '/')
1496 break;
1497
1498 continue;
1499 }
1500 }
1501
1502 errno = ENAMETOOLONG;
1503 if (res + 1 + len + 1 >= tmp1)
1504 return -1;
1505
1506 /* copy one component */
1507 *res = '/';
1508 memcpy (res + 1, beg, len);
1509
1510 /* zero-terminate, for readlink */
1511 res [len + 1] = 0;
1512
1513 /* now check if it's a symlink */
1514 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1515
1516 if (linklen < 0)
741 { 1517 {
742 int len = strlen (name) + 1; 1518 if (errno != EINVAL)
1519 return -1;
743 1520
744 res++; 1521 /* it's a normal directory. hopefully */
745
746 while (memofs + len > memlen)
747 {
748 memlen *= 2;
749 X_LOCK (wrklock);
750 req->ptr2 = names = realloc (names, memlen);
751 X_UNLOCK (wrklock);
752
753 if (!names)
754 break;
755 }
756
757 memcpy (names + memofs, name, len);
758 memofs += len; 1522 res += len + 1;
759 }
760 }
761
762 if (errno)
763 res = -1;
764
765 req->result = res;
766}
767
768/*****************************************************************************/
769
770X_THREAD_PROC (eio_proc)
771{
772 eio_req *req;
773 struct timespec ts;
774 worker *self = (worker *)thr_arg;
775
776 /* try to distribute timeouts somewhat randomly */
777 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778
779 for (;;)
780 {
781 ts.tv_sec = time (0) + IDLE_TIMEOUT;
782
783 X_LOCK (reqlock);
784
785 for (;;)
786 {
787 self->req = req = reqq_shift (&req_queue);
788
789 if (req)
790 break;
791
792 ++idle;
793
794 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
795 {
796 if (idle > max_idle)
797 {
798 --idle;
799 X_UNLOCK (reqlock);
800 X_LOCK (wrklock);
801 --started;
802 X_UNLOCK (wrklock);
803 goto quit;
804 }
805
806 /* we are allowed to idle, so do so without any timeout */
807 X_COND_WAIT (reqwait, reqlock);
808 ts.tv_sec = time (0) + IDLE_TIMEOUT;
809 }
810
811 --idle;
812 }
813
814 --nready;
815
816 X_UNLOCK (reqlock);
817
818 errno = 0; /* strictly unnecessary */
819
820 if (!EIO_CANCELLED (req))
821 switch (req->type)
822 {
823 case EIO_READ: req->result = req->offs >= 0
824 ? pread (req->int1, req->ptr2, req->size, req->offs)
825 : read (req->int1, req->ptr2, req->size); break;
826 case EIO_WRITE: req->result = req->offs >= 0
827 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
828 : write (req->int1, req->ptr2, req->size); break;
829
830 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845 case EIO_CLOSE: req->result = close (req->int1); break;
846 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856 case EIO_SYNC: req->result = 0; sync (); break;
857 case EIO_FSYNC: req->result = fsync (req->int1); break;
858 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860 case EIO_READDIR: scandir_ (req, self); break;
861
862 case EIO_BUSY:
863#ifdef _WIN32
864 Sleep (req->nv1 * 1000.);
865#else
866 {
867 struct timeval tv;
868
869 tv.tv_sec = req->nv1;
870 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872 req->result = select (0, 0, 0, 0, &tv);
873 }
874#endif
875 break;
876
877 case EIO_UTIME:
878 case EIO_FUTIME:
879 {
880 struct timeval tv[2];
881 struct timeval *times;
882
883 if (req->nv1 != -1. || req->nv2 != -1.)
884 {
885 tv[0].tv_sec = req->nv1;
886 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887 tv[1].tv_sec = req->nv2;
888 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890 times = tv;
891 }
892 else
893 times = 0;
894
895
896 req->result = req->type == EIO_FUTIME
897 ? futimes (req->int1, times)
898 : utimes (req->ptr1, times);
899 }
900
901 case EIO_GROUP:
902 case EIO_NOP:
903 break;
904
905 case EIO_QUIT:
906 goto quit;
907
908 default:
909 req->result = -1;
910 break;
911 }
912
913 req->errorno = errno;
914
915 X_LOCK (reslock);
916
917 ++npending;
918
919 if (!reqq_push (&res_queue, req) && want_poll_cb)
920 want_poll_cb ();
921
922 self->req = 0;
923 worker_clear (self);
924
925 X_UNLOCK (reslock);
926 }
927
928quit:
929 X_LOCK (wrklock);
930 worker_free (self);
931 X_UNLOCK (wrklock);
932
933 return 0;
934}
935
936/*****************************************************************************/
937
938static void eio_atfork_prepare (void)
939{
940 X_LOCK (wrklock);
941 X_LOCK (reqlock);
942 X_LOCK (reslock);
943#if !HAVE_PREADWRITE
944 X_LOCK (preadwritelock);
945#endif
946#if !HAVE_READDIR_R
947 X_LOCK (readdirlock);
948#endif
949}
950
951static void eio_atfork_parent (void)
952{
953#if !HAVE_READDIR_R
954 X_UNLOCK (readdirlock);
955#endif
956#if !HAVE_PREADWRITE
957 X_UNLOCK (preadwritelock);
958#endif
959 X_UNLOCK (reslock);
960 X_UNLOCK (reqlock);
961 X_UNLOCK (wrklock);
962}
963
964static void eio_atfork_child (void)
965{
966 eio_req *prv;
967
968 while (prv = reqq_shift (&req_queue))
969 eio_destroy (prv);
970
971 while (prv = reqq_shift (&res_queue))
972 eio_destroy (prv);
973
974 while (wrk_first.next != &wrk_first)
975 {
976 worker *wrk = wrk_first.next;
977
978 if (wrk->req)
979 eio_destroy (wrk->req);
980
981 worker_clear (wrk);
982 worker_free (wrk);
983 }
984
985 started = 0;
986 idle = 0;
987 nreqs = 0;
988 nready = 0;
989 npending = 0;
990
991 eio_atfork_parent ();
992}
993
994int eio_init (void (*want_poll)(void), void (*done_poll)(void))
995{
996 want_poll_cb = want_poll;
997 done_poll_cb = done_poll;
998
999#ifdef _WIN32
1000 X_MUTEX_CHECK (wrklock);
1001 X_MUTEX_CHECK (reslock);
1002 X_MUTEX_CHECK (reqlock);
1003 X_MUTEX_CHECK (reqwait);
1004 X_MUTEX_CHECK (preadwritelock);
1005 X_MUTEX_CHECK (readdirlock);
1006
1007 X_COND_CHECK (reqwait);
1008#endif
1009
1010 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1011}
1012
1013#if 0
1014
1015eio_req *eio_fsync (int fd, eio_cb cb);
1016eio_req *eio_fdatasync (int fd, eio_cb cb);
1017eio_req *eio_dupclose (int fd, eio_cb cb);
1018eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1019eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1020eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1021eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1022eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1023eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1024eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1025eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1026eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1027eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1028eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1029eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1030eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1031eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1032eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1033eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1034eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1035eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1036eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1037eio_req *eio_unlink (const char *path, eio_cb cb);
1038eio_req *eio_rmdir (const char *path, eio_cb cb);
1039eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1040eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1041eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1042eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1043void
1044aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045 PROTOTYPE: $$$;$
1046 PPCODE:
1047{
1048 dREQ;
1049
1050 req->type = EIO_OPEN;
1051 req->sv1 = newSVsv (pathname);
1052 req->ptr1 = SvPVbyte_nolen (req->sv1);
1053 req->int1 = flags;
1054 req->int2 = mode;
1055
1056 EIO_SEND;
1057}
1058
1059void
1060aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061 PROTOTYPE: $;$
1062 ALIAS:
1063 aio_fsync = EIO_FSYNC
1064 aio_fdatasync = EIO_FDATASYNC
1065 PPCODE:
1066{
1067 dREQ;
1068
1069 req->type = ix;
1070 req->sv1 = newSVsv (fh);
1071 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073 EIO_SEND (req);
1074}
1075
1076void
1077aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078 PROTOTYPE: $;$
1079 PPCODE:
1080{
1081 dREQ;
1082
1083 req->type = EIO_CLOSE;
1084 req->sv1 = newSVsv (fh);
1085 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087 EIO_SEND (req);
1088}
1089
1090void
1091aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092 ALIAS:
1093 aio_read = EIO_READ
1094 aio_write = EIO_WRITE
1095 PROTOTYPE: $$$$$;$
1096 PPCODE:
1097{
1098 STRLEN svlen;
1099 char *svptr = SvPVbyte (data, svlen);
1100 UV len = SvUV (length);
1101
1102 SvUPGRADE (data, SVt_PV);
1103 SvPOK_on (data);
1104
1105 if (dataoffset < 0)
1106 dataoffset += svlen;
1107
1108 if (dataoffset < 0 || dataoffset > svlen)
1109 croak ("dataoffset outside of data scalar");
1110
1111 if (ix == EIO_WRITE)
1112 {
1113 /* write: check length and adjust. */
1114 if (!SvOK (length) || len + dataoffset > svlen)
1115 len = svlen - dataoffset;
1116 } 1523 }
1117 else 1524 else
1118 { 1525 {
1119 /* read: grow scalar as necessary */ 1526 /* yay, it was a symlink - build new path in tmp2 */
1120 svptr = SvGROW (data, len + dataoffset + 1); 1527 int rellen = strlen (rel);
1528
1529 errno = ENAMETOOLONG;
1530 if (linklen + 1 + rellen >= PATH_MAX)
1531 return -1;
1532
1533 errno = ELOOP;
1534 if (!--symlinks)
1535 return -1;
1536
1537 if (*tmp1 == '/')
1538 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1539
1540 /* we need to be careful, as rel might point into tmp2 already */
1541 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1542 tmp2 [linklen] = '/';
1543 memcpy (tmp2, tmp1, linklen);
1544
1545 rel = tmp2;
1121 } 1546 }
1547 }
1122 1548
1549 /* special case for the lone root path */
1550 if (res == tmpbuf->ptr)
1551 *res++ = '/';
1552
1553 return res - (char *)tmpbuf->ptr;
1554}
1555
1556static signed char
1557eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1558{
1559 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1560 : a->inode < b->inode ? -1
1561 : a->inode > b->inode ? 1
1562 : 0;
1563}
1564
1565#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1566
1567#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1568#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1569
1570static void
1571eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1572{
1573 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1574 unsigned char *bit = bits;
1575
1576 assert (CHAR_BIT == 8);
1577 assert (sizeof (eio_dirent) * 8 < 256);
1578 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1579 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1580
1581 if (size <= EIO_SORT_FAST)
1582 return;
1583
1584 /* first prepare an array of bits to test in our radix sort */
1585 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1586 /* inode_bits must contain all inodes ORed together */
1587 /* which is used to skip bits that are 0 everywhere, which is very common */
1588 {
1589 eio_ino_t endianness;
1590 int i, j;
1591
1592 /* we store the byte offset of byte n into byte n of "endianness" */
1593 for (i = 0; i < sizeof (eio_ino_t); ++i)
1594 ((unsigned char *)&endianness)[i] = i;
1595
1596 *bit++ = 0;
1597
1598 for (i = 0; i < sizeof (eio_ino_t); ++i)
1599 {
1600 /* shifting off the byte offsets out of "endianness" */
1601 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1602 endianness >>= 8;
1603
1604 for (j = 0; j < 8; ++j)
1605 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1606 *bit++ = offs + j;
1607 }
1608
1609 for (j = 0; j < 8; ++j)
1610 if (score_bits & (1 << j))
1611 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1612 }
1613
1614 /* now actually do the sorting (a variant of MSD radix sort) */
1615 {
1616 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1617 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1618 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1619 int stk_idx = 0;
1620
1621 base_stk [stk_idx] = dents;
1622 end_stk [stk_idx] = dents + size;
1623 bit_stk [stk_idx] = bit - 1;
1624
1625 do
1626 {
1627 base = base_stk [stk_idx];
1628 end = end_stk [stk_idx];
1629 bit = bit_stk [stk_idx];
1630
1631 for (;;)
1632 {
1633 unsigned char O = *bit >> 3;
1634 unsigned char M = 1 << (*bit & 7);
1635
1636 eio_dirent *a = base;
1637 eio_dirent *b = end;
1638
1639 if (b - a < EIO_SORT_CUTOFF)
1640 break;
1641
1642 /* now bit-partition the array on the bit */
1643 /* this ugly asymmetric loop seems to perform much better than typical */
1644 /* partition algos found in the literature */
1645 do
1646 if (!(((unsigned char *)a)[O] & M))
1647 ++a;
1648 else if (!(((unsigned char *)--b)[O] & M))
1649 {
1650 eio_dirent tmp = *a; *a = *b; *b = tmp;
1651 ++a;
1652 }
1653 while (b > a);
1654
1655 /* next bit, or stop, if no bits left in this path */
1656 if (!*--bit)
1657 break;
1658
1659 base_stk [stk_idx] = a;
1660 end_stk [stk_idx] = end;
1661 bit_stk [stk_idx] = bit;
1662 ++stk_idx;
1663
1664 end = a;
1665 }
1666 }
1667 while (stk_idx--);
1668 }
1669}
1670
1671static void
1672eio_dent_insertion_sort (eio_dirent *dents, int size)
1673{
1674 /* first move the smallest element to the front, to act as a sentinel */
1675 {
1676 int i;
1677 eio_dirent *min = dents;
1678
1679 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1680 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1681 if (EIO_DENT_CMP (dents [i], <, *min))
1682 min = &dents [i];
1683
1684 /* swap elements 0 and j (minimum) */
1685 {
1686 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1687 }
1688 }
1689
1690 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1691 {
1692 eio_dirent *i, *j;
1693
1694 for (i = dents + 1; i < dents + size; ++i)
1695 {
1696 eio_dirent value = *i;
1697
1698 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1699 j [1] = j [0];
1700
1701 j [1] = value;
1702 }
1703 }
1704}
1705
1706static void
1707eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1708{
1709 if (size <= 1)
1710 return; /* our insertion sort relies on size > 0 */
1711
1712 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1713 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1714 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1715
1716 /* use an insertion sort at the end, or for small arrays, */
1717 /* as insertion sort is more efficient for small partitions */
1718 eio_dent_insertion_sort (dents, size);
1719}
1720
1721/* read a full directory */
1722static void
1723eio__scandir (eio_req *req, etp_worker *self)
1724{
1725 char *name, *names;
1726 int namesalloc = 4096 - sizeof (void *) * 4;
1727 int namesoffs = 0;
1728 int flags = req->int1;
1729 eio_dirent *dents = 0;
1730 int dentalloc = 128;
1731 int dentoffs = 0;
1732 eio_ino_t inode_bits = 0;
1733#ifdef _WIN32
1734 HANDLE dirp;
1735 WIN32_FIND_DATA entp;
1736#else
1737 DIR *dirp;
1738 EIO_STRUCT_DIRENT *entp;
1739#endif
1740
1741 req->result = -1;
1742
1743 if (!(flags & EIO_READDIR_DENTS))
1744 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1745
1746#ifdef _WIN32
1747 {
1748 int len = strlen ((const char *)req->ptr1);
1749 char *path = malloc (MAX_PATH);
1750 const char *fmt;
1751 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1752
1753 if (!len)
1754 fmt = "./*";
1755 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1756 fmt = "%s*";
1757 else
1758 fmt = "%s/*";
1759
1760 _snprintf (path, MAX_PATH, fmt, reqpath);
1761 dirp = FindFirstFile (path, &entp);
1762 free (path);
1763
1764 if (dirp == INVALID_HANDLE_VALUE)
1765 {
1766 /* should steal _dosmaperr */
1767 switch (GetLastError ())
1768 {
1769 case ERROR_FILE_NOT_FOUND:
1770 req->result = 0;
1771 break;
1772
1773 case ERROR_INVALID_NAME:
1774 case ERROR_PATH_NOT_FOUND:
1775 case ERROR_NO_MORE_FILES:
1776 errno = ENOENT;
1777 break;
1778
1779 case ERROR_NOT_ENOUGH_MEMORY:
1780 errno = ENOMEM;
1781 break;
1782
1783 default:
1784 errno = EINVAL;
1785 break;
1786 }
1787
1788 return;
1789 }
1790 }
1791#else
1792 #if HAVE_AT
1793 if (req->wd)
1794 {
1795 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1796
1123 if (len < 0) 1797 if (fd < 0)
1124 croak ("length must not be negative"); 1798 return;
1125 1799
1800 dirp = fdopendir (fd);
1801
1802 if (!dirp)
1803 close (fd);
1804 }
1805 else
1806 dirp = opendir (req->ptr1);
1807 #else
1808 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1809 #endif
1810
1811 if (!dirp)
1812 return;
1813#endif
1814
1815 if (req->flags & EIO_FLAG_PTR1_FREE)
1816 free (req->ptr1);
1817
1818 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1819 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1820 req->ptr2 = names = malloc (namesalloc);
1821
1822 if (!names || (flags && !dents))
1823 return;
1824
1825 for (;;)
1826 {
1827 int done;
1828
1829#ifdef _WIN32
1830 done = !dirp;
1831#else
1832 errno = 0;
1833 entp = readdir (dirp);
1834 done = !entp;
1835#endif
1836
1837 if (done)
1126 { 1838 {
1127 dREQ; 1839#ifndef _WIN32
1840 int old_errno = errno;
1841 closedir (dirp);
1842 errno = old_errno;
1128 1843
1129 req->type = ix; 1844 if (errno)
1130 req->sv1 = newSVsv (fh); 1845 break;
1131 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1846#endif
1132 : IoOFP (sv_2io (fh)));
1133 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134 req->size = len;
1135 req->sv2 = SvREFCNT_inc (data);
1136 req->ptr2 = (char *)svptr + dataoffset;
1137 req->stroffset = dataoffset;
1138 1847
1139 if (!SvREADONLY (data)) 1848 /* sort etc. */
1849 req->int1 = flags;
1850 req->result = dentoffs;
1851
1852 if (flags & EIO_READDIR_STAT_ORDER)
1853 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1854 else if (flags & EIO_READDIR_DIRS_FIRST)
1855 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1856 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1857 else
1858 {
1859 /* in this case, all is known, and we just put dirs first and sort them */
1860 eio_dirent *oth = dents + dentoffs;
1861 eio_dirent *dir = dents;
1862
1863 /* now partition dirs to the front, and non-dirs to the back */
1864 /* by walking from both sides and swapping if necessary */
1865 while (oth > dir)
1866 {
1867 if (dir->type == EIO_DT_DIR)
1868 ++dir;
1869 else if ((--oth)->type == EIO_DT_DIR)
1870 {
1871 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1872
1873 ++dir;
1874 }
1875 }
1876
1877 /* now sort the dirs only (dirs all have the same score) */
1878 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1879 }
1880
1881 break;
1882 }
1883
1884 /* now add the entry to our list(s) */
1885 name = D_NAME (entp);
1886
1887 /* skip . and .. entries */
1888 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1889 {
1890 int len = D_NAMLEN (entp) + 1;
1891
1892 while (ecb_expect_false (namesoffs + len > namesalloc))
1140 { 1893 {
1141 SvREADONLY_on (data); 1894 namesalloc *= 2;
1142 req->flags |= FLAG_SV2_RO_OFF; 1895 req->ptr2 = names = realloc (names, namesalloc);
1896
1897 if (!names)
1898 break;
1143 } 1899 }
1144 1900
1145 EIO_SEND; 1901 memcpy (names + namesoffs, name, len);
1902
1903 if (dents)
1904 {
1905 struct eio_dirent *ent;
1906
1907 if (ecb_expect_false (dentoffs == dentalloc))
1908 {
1909 dentalloc *= 2;
1910 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1911
1912 if (!dents)
1913 break;
1914 }
1915
1916 ent = dents + dentoffs;
1917
1918 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1919 ent->namelen = len - 1;
1920 ent->inode = D_INO (entp);
1921
1922 inode_bits |= ent->inode;
1923
1924 switch (D_TYPE (entp))
1925 {
1926 default:
1927 ent->type = EIO_DT_UNKNOWN;
1928 flags |= EIO_READDIR_FOUND_UNKNOWN;
1929 break;
1930
1931 #ifdef DT_FIFO
1932 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1933 #endif
1934 #ifdef DT_CHR
1935 case DT_CHR: ent->type = EIO_DT_CHR; break;
1936 #endif
1937 #ifdef DT_MPC
1938 case DT_MPC: ent->type = EIO_DT_MPC; break;
1939 #endif
1940 #ifdef DT_DIR
1941 case DT_DIR: ent->type = EIO_DT_DIR; break;
1942 #endif
1943 #ifdef DT_NAM
1944 case DT_NAM: ent->type = EIO_DT_NAM; break;
1945 #endif
1946 #ifdef DT_BLK
1947 case DT_BLK: ent->type = EIO_DT_BLK; break;
1948 #endif
1949 #ifdef DT_MPB
1950 case DT_MPB: ent->type = EIO_DT_MPB; break;
1951 #endif
1952 #ifdef DT_REG
1953 case DT_REG: ent->type = EIO_DT_REG; break;
1954 #endif
1955 #ifdef DT_NWK
1956 case DT_NWK: ent->type = EIO_DT_NWK; break;
1957 #endif
1958 #ifdef DT_CMP
1959 case DT_CMP: ent->type = EIO_DT_CMP; break;
1960 #endif
1961 #ifdef DT_LNK
1962 case DT_LNK: ent->type = EIO_DT_LNK; break;
1963 #endif
1964 #ifdef DT_SOCK
1965 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1966 #endif
1967 #ifdef DT_DOOR
1968 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1969 #endif
1970 #ifdef DT_WHT
1971 case DT_WHT: ent->type = EIO_DT_WHT; break;
1972 #endif
1973 }
1974
1975 ent->score = 7;
1976
1977 if (flags & EIO_READDIR_DIRS_FIRST)
1978 {
1979 if (ent->type == EIO_DT_UNKNOWN)
1980 {
1981 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1982 ent->score = 1;
1983 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1984 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1985 }
1986 else if (ent->type == EIO_DT_DIR)
1987 ent->score = 0;
1988 }
1989 }
1990
1991 namesoffs += len;
1992 ++dentoffs;
1146 } 1993 }
1994
1995 if (EIO_CANCELLED (req))
1996 {
1997 errno = ECANCELED;
1998 break;
1999 }
2000
2001#ifdef _WIN32
2002 if (!FindNextFile (dirp, &entp))
2003 {
2004 FindClose (dirp);
2005 dirp = 0;
2006 }
2007#endif
2008 }
2009}
2010
2011/*****************************************************************************/
2012/* working directory stuff */
2013/* various deficiencies in the posix 2008 api force us to */
2014/* keep the absolute path in string form at all times */
2015/* fuck yeah. */
2016
2017#if !HAVE_AT
2018
2019/* a bit like realpath, but usually faster because it doesn'T have to return */
2020/* an absolute or canonical path */
2021static const char *
2022wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2023{
2024 if (!wd || *path == '/')
2025 return path;
2026
2027 if (path [0] == '.' && !path [1])
2028 return wd->str;
2029
2030 {
2031 int l1 = wd->len;
2032 int l2 = strlen (path);
2033
2034 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2035
2036 memcpy (res, wd->str, l1);
2037 res [l1] = '/';
2038 memcpy (res + l1 + 1, path, l2 + 1);
2039
2040 return res;
2041 }
2042}
2043
2044#endif
2045
2046static eio_wd
2047eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2048{
2049 int fd;
2050 eio_wd res;
2051 int len = eio__realpath (tmpbuf, wd, path);
2052
2053 if (len < 0)
2054 return EIO_INVALID_WD;
2055
2056#if HAVE_AT
2057 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2058
2059 if (fd < 0)
2060 return EIO_INVALID_WD;
2061#endif
2062
2063 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2064
2065#if HAVE_AT
2066 res->fd = fd;
2067#endif
2068
2069 res->len = len;
2070 memcpy (res->str, tmpbuf->ptr, len);
2071 res->str [len] = 0;
2072
2073 return res;
2074}
2075
2076eio_wd
2077eio_wd_open_sync (eio_wd wd, const char *path)
2078{
2079 struct tmpbuf tmpbuf = { 0 };
2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
2082
2083 return wd;
1147} 2084}
1148 2085
1149void 2086void
1150aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 2087eio_wd_close_sync (eio_wd wd)
1151 PROTOTYPE: $$;$
1152 PPCODE:
1153{ 2088{
1154 SV *data; 2089 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1155 dREQ; 2090 {
1156 2091 #if HAVE_AT
1157 data = newSV (NAME_MAX); 2092 close (wd->fd);
1158 SvPOK_on (data); 2093 #endif
1159 2094 free (wd);
1160 req->type = EIO_READLINK; 2095 }
1161 req->sv1 = newSVsv (path);
1162 req->ptr1 = SvPVbyte_nolen (req->sv1);
1163 req->sv2 = data;
1164 req->ptr2 = SvPVbyte_nolen (data);
1165
1166 EIO_SEND;
1167} 2096}
2097
2098#if HAVE_AT
2099
2100/* they forgot these */
2101
2102static int
2103eio__truncateat (int dirfd, const char *path, off_t length)
2104{
2105 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = ftruncate (fd, length);
2112 close (fd);
2113 return res;
2114}
2115
2116static int
2117eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2118{
2119 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2120 int res;
2121
2122 if (fd < 0)
2123 return fd;
2124
2125 res = fstatvfs (fd, buf);
2126 close (fd);
2127 return res;
2128
2129}
2130
2131#endif
2132
2133/*****************************************************************************/
2134
2135#define ALLOC(len) \
2136 if (!req->ptr2) \
2137 { \
2138 X_LOCK (wrklock); \
2139 req->flags |= EIO_FLAG_PTR2_FREE; \
2140 X_UNLOCK (wrklock); \
2141 req->ptr2 = malloc (len); \
2142 if (!req->ptr2) \
2143 { \
2144 errno = ENOMEM; \
2145 req->result = -1; \
2146 break; \
2147 } \
2148 }
2149
2150static void ecb_noinline ecb_cold
2151etp_proc_init (void)
2152{
2153#if HAVE_PRCTL_SET_NAME
2154 /* provide a more sensible "thread name" */
2155 char name[16 + 1];
2156 const int namelen = sizeof (name) - 1;
2157 int len;
2158
2159 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2160 name [namelen] = 0;
2161 len = strlen (name);
2162 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2163 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2164#endif
2165}
2166
2167X_THREAD_PROC (etp_proc)
2168{
2169 ETP_REQ *req;
2170 struct timespec ts;
2171 etp_worker *self = (etp_worker *)thr_arg;
2172
2173 etp_proc_init ();
2174
2175 /* try to distribute timeouts somewhat evenly */
2176 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
2177
2178 for (;;)
2179 {
2180 ts.tv_sec = 0;
2181
2182 X_LOCK (reqlock);
2183
2184 for (;;)
2185 {
2186 req = reqq_shift (&req_queue);
2187
2188 if (req)
2189 break;
2190
2191 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2192 {
2193 X_UNLOCK (reqlock);
2194 X_LOCK (wrklock);
2195 --started;
2196 X_UNLOCK (wrklock);
2197 goto quit;
2198 }
2199
2200 ++idle;
2201
2202 if (idle <= max_idle)
2203 /* we are allowed to idle, so do so without any timeout */
2204 X_COND_WAIT (reqwait, reqlock);
2205 else
2206 {
2207 /* initialise timeout once */
2208 if (!ts.tv_sec)
2209 ts.tv_sec = time (0) + idle_timeout;
2210
2211 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2212 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
2213 }
2214
2215 --idle;
2216 }
2217
2218 --nready;
2219
2220 X_UNLOCK (reqlock);
2221
2222 if (req->type < 0)
2223 goto quit;
2224
2225 ETP_EXECUTE (self, req);
2226
2227 X_LOCK (reslock);
2228
2229 ++npending;
2230
2231 if (!reqq_push (&res_queue, req) && want_poll_cb)
2232 want_poll_cb ();
2233
2234 etp_worker_clear (self);
2235
2236 X_UNLOCK (reslock);
2237 }
2238
2239quit:
2240 free (req);
2241
2242 X_LOCK (wrklock);
2243 etp_worker_free (self);
2244 X_UNLOCK (wrklock);
2245}
2246
2247/*****************************************************************************/
2248
2249int ecb_cold
2250eio_init (void (*want_poll)(void), void (*done_poll)(void))
2251{
2252#if !HAVE_PREADWRITE
2253 X_MUTEX_CREATE (preadwritelock);
2254#endif
2255
2256 return etp_init (want_poll, done_poll);
2257}
2258
2259ecb_inline void
2260eio_api_destroy (eio_req *req)
2261{
2262 free (req);
2263}
2264
2265#define REQ(rtype) \
2266 eio_req *req; \
2267 \
2268 req = (eio_req *)calloc (1, sizeof *req); \
2269 if (!req) \
2270 return 0; \
2271 \
2272 req->type = rtype; \
2273 req->pri = pri; \
2274 req->finish = cb; \
2275 req->data = data; \
2276 req->destroy = eio_api_destroy;
2277
2278#define SEND eio_submit (req); return req
2279
2280#define PATH \
2281 req->flags |= EIO_FLAG_PTR1_FREE; \
2282 req->ptr1 = strdup (path); \
2283 if (!req->ptr1) \
2284 { \
2285 eio_api_destroy (req); \
2286 return 0; \
2287 }
2288
2289static void
2290eio_execute (etp_worker *self, eio_req *req)
2291{
2292#if HAVE_AT
2293 int dirfd;
2294#else
2295 const char *path;
2296#endif
2297
2298 if (ecb_expect_false (EIO_CANCELLED (req)))
2299 {
2300 req->result = -1;
2301 req->errorno = ECANCELED;
2302 return;
2303 }
2304
2305 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2306 {
2307 req->result = -1;
2308 req->errorno = ENOENT;
2309 return;
2310 }
2311
2312 if (req->type >= EIO_OPEN)
2313 {
2314 #if HAVE_AT
2315 dirfd = WD2FD (req->wd);
2316 #else
2317 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2318 #endif
2319 }
2320
2321 switch (req->type)
2322 {
2323 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2324 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2325 break;
2326 case EIO_WD_CLOSE: req->result = 0;
2327 eio_wd_close_sync (req->wd); break;
2328
2329 case EIO_SEEK: eio__lseek (req); break;
2330 case EIO_READ: ALLOC (req->size);
2331 req->result = req->offs >= 0
2332 ? pread (req->int1, req->ptr2, req->size, req->offs)
2333 : read (req->int1, req->ptr2, req->size); break;
2334 case EIO_WRITE: req->result = req->offs >= 0
2335 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2336 : write (req->int1, req->ptr2, req->size); break;
2337
2338 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2339 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2340
2341#if HAVE_AT
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2347 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2348 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2349 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2350 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2353 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2354 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2445
2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2448
2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2452
2453 case EIO_CLOSE: req->result = close (req->int1); break;
2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2455 case EIO_SYNC: req->result = 0; sync (); break;
2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2465
2466 case EIO_READDIR: eio__scandir (req, self); break;
2467
2468 case EIO_BUSY:
2469#ifdef _WIN32
2470 Sleep (req->nv1 * 1e3);
2471#else
2472 {
2473 struct timeval tv;
2474
2475 tv.tv_sec = req->nv1;
2476 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2477
2478 req->result = select (0, 0, 0, 0, &tv);
2479 }
2480#endif
2481 break;
2482
2483 case EIO_GROUP:
2484 abort (); /* handled in eio_request */
2485
2486 case EIO_NOP:
2487 req->result = 0;
2488 break;
2489
2490 case EIO_CUSTOM:
2491 req->feed (req);
2492 break;
2493
2494 default:
2495 req->result = EIO_ENOSYS ();
2496 break;
2497 }
2498
2499 req->errorno = errno;
2500}
2501
2502#ifndef EIO_NO_WRAPPERS
2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2515{
2516 REQ (EIO_NOP); SEND;
2517}
2518
2519eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2520{
2521 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2522}
2523
2524eio_req *eio_sync (int pri, eio_cb cb, void *data)
2525{
2526 REQ (EIO_SYNC); SEND;
2527}
2528
2529eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2530{
2531 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2532}
2533
2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2537}
2538
2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2542}
2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2555{
2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2557}
2558
2559eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2560{
2561 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2562}
2563
2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2565{
2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2567}
2568
2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2572}
2573
2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2575{
2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2577}
2578
2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2580{
2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2582}
2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2590{
2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2592}
2593
2594eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2595{
2596 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2597}
2598
2599eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2600{
2601 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2602}
2603
2604eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2605{
2606 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2607}
2608
2609eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2610{
2611 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2612}
2613
2614eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2615{
2616 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2617}
2618
2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2620{
2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2622}
2623
2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2625{
2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2627}
2628
2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2630{
2631 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2632}
2633
2634eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2635{
2636 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2637}
2638
2639eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2640{
2641 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2642}
2643
2644eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2645{
2646 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2647}
2648
2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2650{
2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2652}
2653
2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2655{
2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2657}
2658
2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2660{
2661 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2662}
2663
2664eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2665{
2666 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2667}
2668
2669static eio_req *
2670eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2671{
2672 REQ (type); PATH; SEND;
2673}
2674
2675eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2676{
2677 return eio__1path (EIO_READLINK, path, pri, cb, data);
2678}
2679
2680eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2681{
2682 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2683}
2684
2685eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2686{
2687 return eio__1path (EIO_STAT, path, pri, cb, data);
2688}
2689
2690eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2691{
2692 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2693}
2694
2695eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2696{
2697 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2698}
2699
2700eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2701{
2702 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2703}
2704
2705eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2706{
2707 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2708}
2709
2710eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2711{
2712 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2713}
2714
2715eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2716{
2717 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2718}
2719
2720static eio_req *
2721eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2722{
2723 REQ (type); PATH;
2724
2725 req->flags |= EIO_FLAG_PTR2_FREE;
2726 req->ptr2 = strdup (new_path);
2727 if (!req->ptr2)
2728 {
2729 eio_api_destroy (req);
2730 return 0;
2731 }
2732
2733 SEND;
2734}
2735
2736eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2737{
2738 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2739}
2740
2741eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2742{
2743 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2744}
2745
2746eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2747{
2748 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2749}
2750
2751eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2752{
2753 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2754}
2755
2756#endif
2757
2758eio_req *eio_grp (eio_cb cb, void *data)
2759{
2760 const int pri = EIO_PRI_MAX;
2761
2762 REQ (EIO_GROUP); SEND;
2763}
2764
2765#undef REQ
2766#undef PATH
2767#undef SEND
2768
2769/*****************************************************************************/
2770/* grp functions */
1168 2771
1169void 2772void
1170aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171 PROTOTYPE: $$$$;$
1172 PPCODE:
1173{
1174 dREQ;
1175
1176 req->type = EIO_SENDFILE;
1177 req->sv1 = newSVsv (out_fh);
1178 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179 req->sv2 = newSVsv (in_fh);
1180 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181 req->offs = SvVAL64 (in_offset);
1182 req->size = length;
1183
1184 EIO_SEND;
1185}
1186
1187void
1188aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189 PROTOTYPE: $$$;$
1190 PPCODE:
1191{
1192 dREQ;
1193
1194 req->type = EIO_READAHEAD;
1195 req->sv1 = newSVsv (fh);
1196 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197 req->offs = SvVAL64 (offset);
1198 req->size = length;
1199
1200 EIO_SEND;
1201}
1202
1203void
1204aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205 ALIAS:
1206 aio_stat = EIO_STAT
1207 aio_lstat = EIO_LSTAT
1208 PPCODE:
1209{
1210 dREQ;
1211
1212 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213 if (!req->ptr2)
1214 {
1215 req_destroy (req);
1216 croak ("out of memory during aio_stat statdata allocation");
1217 }
1218
1219 req->flags |= FLAG_PTR2_FREE;
1220 req->sv1 = newSVsv (fh_or_path);
1221
1222 if (SvPOK (fh_or_path))
1223 {
1224 req->type = ix;
1225 req->ptr1 = SvPVbyte_nolen (req->sv1);
1226 }
1227 else
1228 {
1229 req->type = EIO_FSTAT;
1230 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231 }
1232
1233 EIO_SEND;
1234}
1235
1236void
1237aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238 PPCODE:
1239{
1240 dREQ;
1241
1242 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244 req->sv1 = newSVsv (fh_or_path);
1245
1246 if (SvPOK (fh_or_path))
1247 {
1248 req->type = EIO_UTIME;
1249 req->ptr1 = SvPVbyte_nolen (req->sv1);
1250 }
1251 else
1252 {
1253 req->type = EIO_FUTIME;
1254 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255 }
1256
1257 EIO_SEND;
1258}
1259
1260void
1261aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262 PPCODE:
1263{
1264 dREQ;
1265
1266 req->sv1 = newSVsv (fh_or_path);
1267 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269 if (SvPOK (fh_or_path))
1270 {
1271 req->type = EIO_TRUNCATE;
1272 req->ptr1 = SvPVbyte_nolen (req->sv1);
1273 }
1274 else
1275 {
1276 req->type = EIO_FTRUNCATE;
1277 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278 }
1279
1280 EIO_SEND;
1281}
1282
1283void
1284aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285 ALIAS:
1286 aio_chmod = EIO_CHMOD
1287 aio_fchmod = EIO_FCHMOD
1288 aio_mkdir = EIO_MKDIR
1289 PPCODE:
1290{
1291 dREQ;
1292
1293 req->type = type;
1294 req->int2 = mode;
1295 req->sv1 = newSVsv (fh_or_path);
1296
1297 if (ix == EIO_FCHMOD)
1298 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299 else
1300 req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302 EIO_SEND;
1303}
1304
1305void
1306aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307 PPCODE:
1308{
1309 dREQ;
1310
1311 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313 req->sv1 = newSVsv (fh_or_path);
1314
1315 if (SvPOK (fh_or_path))
1316 {
1317 req->type = EIO_CHOWN;
1318 req->ptr1 = SvPVbyte_nolen (req->sv1);
1319 }
1320 else
1321 {
1322 req->type = EIO_FCHOWN;
1323 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324 }
1325
1326 EIO_SEND;
1327}
1328
1329void
1330aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331 ALIAS:
1332 aio_unlink = EIO_UNLINK
1333 aio_rmdir = EIO_RMDIR
1334 aio_readdir = EIO_READDIR
1335 PPCODE:
1336{
1337 dREQ;
1338
1339 req->type = ix;
1340 req->sv1 = newSVsv (pathname);
1341 req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343 EIO_SEND;
1344}
1345
1346void
1347aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348 ALIAS:
1349 aio_link = EIO_LINK
1350 aio_symlink = EIO_SYMLINK
1351 aio_rename = EIO_RENAME
1352 PPCODE:
1353{
1354 dREQ;
1355
1356 req->type = ix;
1357 req->sv1 = newSVsv (oldpath);
1358 req->ptr1 = SvPVbyte_nolen (req->sv1);
1359 req->sv2 = newSVsv (newpath);
1360 req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362 EIO_SEND;
1363}
1364
1365void
1366aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367 PPCODE:
1368{
1369 dREQ;
1370
1371 req->type = EIO_MKNOD;
1372 req->sv1 = newSVsv (pathname);
1373 req->ptr1 = SvPVbyte_nolen (req->sv1);
1374 req->int2 = (mode_t)mode;
1375 req->offs = dev;
1376
1377 EIO_SEND;
1378}
1379
1380void
1381aio_busy (double delay, SV *callback=&PL_sv_undef)
1382 PPCODE:
1383{
1384 dREQ;
1385
1386 req->type = EIO_BUSY;
1387 req->nv1 = delay < 0. ? 0. : delay;
1388
1389 EIO_SEND;
1390}
1391
1392void
1393aio_group (SV *callback=&PL_sv_undef)
1394 PROTOTYPE: ;$
1395 PPCODE:
1396{
1397 dREQ;
1398
1399 req->type = EIO_GROUP;
1400
1401 req_send (req);
1402 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403}
1404
1405void
1406aio_nop (SV *callback=&PL_sv_undef)
1407 ALIAS:
1408 aio_nop = EIO_NOP
1409 aio_sync = EIO_SYNC
1410 PPCODE:
1411{
1412 dREQ;
1413
1414#endif
1415
1416void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2773eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417{ 2774{
1418 grp->int2 = limit; 2775 grp->int2 = limit;
1419 grp->feed = feed; 2776 grp->feed = feed;
1420 2777
1421 grp_try_feed (grp); 2778 grp_try_feed (grp);
1422} 2779}
1423 2780
2781void
1424void eio_grp_limit (eio_req *grp, int limit) 2782eio_grp_limit (eio_req *grp, int limit)
1425{ 2783{
1426 grp->int2 = limit; 2784 grp->int2 = limit;
1427 2785
1428 grp_try_feed (grp); 2786 grp_try_feed (grp);
1429} 2787}
1430 2788
2789void
1431void eio_grp_add (eio_req *grp, eio_req *req) 2790eio_grp_add (eio_req *grp, eio_req *req)
1432{ 2791{
1433 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2792 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2793
2794 grp->flags |= EIO_FLAG_GROUPADD;
1434 2795
1435 ++grp->size; 2796 ++grp->size;
1436 req->grp = grp; 2797 req->grp = grp;
1437 2798
1438 req->grp_prev = 0; 2799 req->grp_prev = 0;
… …
1442 grp->grp_first->grp_prev = req; 2803 grp->grp_first->grp_prev = req;
1443 2804
1444 grp->grp_first = req; 2805 grp->grp_first = req;
1445} 2806}
1446 2807
2808/*****************************************************************************/
2809/* misc garbage */
1447 2810
2811eio_ssize_t
2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2813{
2814 return eio__sendfile (ofd, ifd, offset, count);
2815}
2816

Diff Legend

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