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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.117 by root, Sun Apr 1 17:22:42 2012 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
18 81
19#ifndef EIO_FINISH 82#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 84#endif
22 85
… …
26 89
27#ifndef EIO_FEED 90#ifndef EIO_FEED
28# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
29#endif 92#endif
30 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
31#ifdef _WIN32 105#ifdef _WIN32
32 106
33 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 196
35#else 197#else
36 198
37# include "config.h"
38# include <sys/time.h> 199 #include <sys/time.h>
39# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
40# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
41# include <utime.h> 237# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 238#endif
47 239
48#endif 240#if HAVE_SYS_SYSCALL_H
49 241# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
246#endif
53 247
54#if HAVE_SENDFILE 248#if HAVE_SENDFILE
55# if __linux 249# if __linux
56# include <sys/sendfile.h> 250# include <sys/sendfile.h>
57# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 252# include <sys/socket.h>
59# include <sys/uio.h> 253# include <sys/uio.h>
60# elif __hpux 254# elif __hpux
61# include <sys/socket.h> 255# include <sys/socket.h>
62# elif __solaris /* not yet */ 256# elif __solaris
63# include <sys/sendfile.h> 257# include <sys/sendfile.h>
64# else 258# else
65# error sendfile support requested but not available 259# error sendfile support requested but not available
66# endif 260# endif
67#endif 261#endif
68 262
69/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
71 272
72/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 274#ifndef NAME_MAX
74# define NAME_MAX 4096 275# define NAME_MAX 4096
75#endif 276#endif
76 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
77/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
79 285
80#define dBUF \ 286#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 288 errno = ENOMEM; \
85 if (!eio_buf) \ 289 if (!eio_buf) \
86 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
87 294
88#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
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;
89 366
90static void (*want_poll_cb) (void); 367static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
92 369
93static unsigned int max_poll_time = 0; 370static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 371static unsigned int max_poll_reqs; /* reslock */
95 372
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 373static unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 374static unsigned int nready; /* reqlock */
98{ 375static unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 378
103static 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)
104 411
105/* worker threads management */ 412/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 413
108typedef struct worker { 414static void
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 415etp_worker_clear (etp_worker *wrk)
123{ 416{
124 if (wrk->dirp) 417}
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129 418
130 if (wrk->dbuf) 419static void ecb_cold
131 { 420etp_worker_free (etp_worker *wrk)
421{
132 free (wrk->dbuf); 422 free (wrk->tmpbuf.ptr);
133 wrk->dbuf = 0;
134 }
135}
136 423
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
141 426
142 free (wrk); 427 free (wrk);
143} 428}
144 429
145static volatile unsigned int nreqs, nready, npending; 430static unsigned int
146static volatile unsigned int max_idle = 4; 431etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 432{
433 int retval;
434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
435 retval = nreqs;
436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 437 return retval;
155} 438}
156 439
157unsigned int eio_nready (void) 440static unsigned int
441etp_nready (void)
158{ 442{
159 unsigned int retval; 443 unsigned int retval;
160 444
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 446 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 448
165 return retval; 449 return retval;
166} 450}
167 451
168unsigned int eio_npending (void) 452static unsigned int
453etp_npending (void)
169{ 454{
170 unsigned int retval; 455 unsigned int retval;
171 456
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 458 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 460
176 return retval; 461 return retval;
177} 462}
178 463
179unsigned int eio_nthreads (void) 464static unsigned int
465etp_nthreads (void)
180{ 466{
181 unsigned int retval; 467 unsigned int retval;
182 468
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 470 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 477 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 478 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 479 * per shift, the most expensive operation.
194 */ 480 */
195typedef struct { 481typedef struct {
196 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 */
197 int size; 483 int size;
198} reqq; 484} etp_reqq;
199 485
200static reqq req_queue; 486static etp_reqq req_queue;
201static reqq res_queue; 487static etp_reqq res_queue;
202 488
203static 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)
204{ 502{
205 int pri = req->pri; 503 int pri = req->pri;
206 req->next = 0; 504 req->next = 0;
207 505
208 if (q->qe[pri]) 506 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
215 513
216 return q->size++; 514 return q->size++;
217} 515}
218 516
219static eio_req *reqq_shift (reqq *q) 517static ETP_REQ * ecb_noinline
518reqq_shift (etp_reqq *q)
220{ 519{
221 int pri; 520 int pri;
222 521
223 if (!q->size) 522 if (!q->size)
224 return 0; 523 return 0;
225 524
226 --q->size; 525 --q->size;
227 526
228 for (pri = EIO_NUM_PRI; pri--; ) 527 for (pri = ETP_NUM_PRI; pri--; )
229 { 528 {
230 eio_req *req = q->qs[pri]; 529 eio_req *req = q->qs[pri];
231 530
232 if (req) 531 if (req)
233 { 532 {
… …
239 } 538 }
240 539
241 abort (); 540 abort ();
242} 541}
243 542
244static void grp_try_feed (eio_req *grp) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 545{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 546 X_MUTEX_CREATE (wrklock);
247 { 547 X_MUTEX_CREATE (reslock);
248 int old_len = grp->size; 548 X_MUTEX_CREATE (reqlock);
549 X_COND_CREATE (reqwait);
249 550
250 EIO_FEED (grp); 551 reqq_init (&req_queue);
552 reqq_init (&res_queue);
251 553
252 /* stop if no progress has been made */ 554 wrk_first.next =
253 if (old_len == grp->size) 555 wrk_first.prev = &wrk_first;
254 {
255 grp->feed = 0;
256 break;
257 }
258 }
259}
260 556
261static int eio_finish (eio_req *req); 557 started = 0;
558 idle = 0;
559 nreqs = 0;
560 nready = 0;
561 npending = 0;
262 562
263static int grp_dec (eio_req *grp) 563 want_poll_cb = want_poll;
264{ 564 done_poll_cb = done_poll;
265 --grp->size;
266 565
267 /* call feeder, if applicable */
268 grp_try_feed (grp);
269
270 /* finish, if done */
271 if (!grp->size && grp->int1)
272 return eio_finish (grp);
273 else
274 return 0; 566 return 0;
275} 567}
276 568
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 569X_THREAD_PROC (etp_proc);
326 570
327static void start_thread (void) 571static void ecb_cold
572etp_start_thread (void)
328{ 573{
329 worker *wrk = calloc (1, sizeof (worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 575
331 if (!wrk) 576 /*TODO*/
332 croak ("unable to allocate worker thread data"); 577 assert (("unable to allocate worker thread data", wrk));
333 578
334 X_LOCK (wrklock); 579 X_LOCK (wrklock);
335 580
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 581 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 582 {
338 wrk->prev = &wrk_first; 583 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 584 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 585 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 586 wrk_first.next = wrk;
… …
345 free (wrk); 590 free (wrk);
346 591
347 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
348} 593}
349 594
595static void
350static void maybe_start_thread (void) 596etp_maybe_start_thread (void)
351{ 597{
352 if (eio_nthreads () >= wanted) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 599 return;
354 600
355 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
356 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 ()))
357 return; 603 return;
358 604
359 start_thread (); 605 etp_start_thread ();
360} 606}
361 607
362void eio_submit (eio_req *req) 608static void ecb_cold
609etp_end_thread (void)
363{ 610{
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{
377 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
378 612
379 req->type = EIO_QUIT; 613 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 615
382 X_LOCK (reqlock); 616 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 621 X_LOCK (wrklock);
388 --started; 622 --started;
389 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
390} 624}
391 625
392void eio_set_max_poll_time (double nseconds) 626static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 627etp_poll (void)
429{ 628{
430 int maxreqs = max_poll_reqs; 629 unsigned int maxreqs;
630 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
432 eio_req *req;
433 632
633 X_LOCK (reslock);
634 maxreqs = max_poll_reqs;
635 maxtime = max_poll_time;
636 X_UNLOCK (reslock);
637
434 if (max_poll_time) 638 if (maxtime)
435 gettimeofday (&tv_start, 0); 639 gettimeofday (&tv_start, 0);
436 640
437 for (;;) 641 for (;;)
438 { 642 {
643 ETP_REQ *req;
644
439 maybe_start_thread (); 645 etp_maybe_start_thread ();
440 646
441 X_LOCK (reslock); 647 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 648 req = reqq_shift (&res_queue);
443 649
444 if (req) 650 if (req)
… …
452 X_UNLOCK (reslock); 658 X_UNLOCK (reslock);
453 659
454 if (!req) 660 if (!req)
455 return 0; 661 return 0;
456 662
663 X_LOCK (reqlock);
457 --nreqs; 664 --nreqs;
665 X_UNLOCK (reqlock);
458 666
459 if (req->type == EIO_GROUP && req->size) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 668 {
461 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
462 continue; 670 continue;
463 } 671 }
464 else 672 else
465 { 673 {
466 int res = eio_finish (req); 674 int res = ETP_FINISH (req);
467 if (res) 675 if (ecb_expect_false (res))
468 return res; 676 return res;
469 } 677 }
470 678
471 if (maxreqs && !--maxreqs) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 680 break;
473 681
474 if (max_poll_time) 682 if (maxtime)
475 { 683 {
476 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
477 685
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 686 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 687 break;
480 } 688 }
481 } 689 }
482 690
483 errno = EAGAIN; 691 errno = EAGAIN;
484 return -1; 692 return -1;
485} 693}
486 694
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
487/*****************************************************************************/ 945/*****************************************************************************/
488/* work around various missing functions */ 946/* work around various missing functions */
489 947
490#if !HAVE_PREADWRITE 948#if !HAVE_PREADWRITE
949# undef pread
950# undef pwrite
491# define pread aio_pread 951# define pread eio__pread
492# define pwrite aio_pwrite 952# define pwrite eio__pwrite
493 953
494/* 954static eio_ssize_t
495 * make our pread/pwrite safe against themselves, but not against 955eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 956{
503 ssize_t res; 957 eio_ssize_t res;
504 off_t ooffset; 958 off_t ooffset;
505 959
506 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
511 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
512 966
513 return res; 967 return res;
514} 968}
515 969
516static 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)
517{ 972{
518 ssize_t res; 973 eio_ssize_t res;
519 off_t ooffset; 974 off_t ooffset;
520 975
521 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 979 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 980 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 981 X_UNLOCK (preadwritelock);
527 982
528 return res; 983 return res;
529} 984}
530#endif 985#endif
531 986
532#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
533 988
989# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 991
992static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
538{ 994{
539 if (times) 995 if (times)
540 { 996 {
541 struct utimbuf buf; 997 struct utimbuf buf;
542 998
… …
547 } 1003 }
548 else 1004 else
549 return utime (filename, 0); 1005 return utime (filename, 0);
550} 1006}
551 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
553{ 1017{
554 errno = ENOSYS; 1018 errno = ENOSYS;
555 return -1; 1019 return -1;
556} 1020}
557 1021
558#endif 1022#endif
559 1023
560#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
1025# undef fdatasync
561# define fdatasync fsync 1026# define fdatasync(fd) fsync (fd)
562#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 = -1;
1038 errno = ENOSYS;
1039#endif
1040
1041 if (res < 0 && errno == ENOSYS && fd >= 0)
1042 sync ();
1043
1044 return res;
1045}
1046
1047/* sync_file_range always needs emulation */
1048static int
1049eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1050{
1051#if HAVE_SYNC_FILE_RANGE
1052 int res;
1053
1054 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1055 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1056 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1057 {
1058 flags = 0
1059 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1060 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1061 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1062 }
1063
1064 res = sync_file_range (fd, offset, nbytes, flags);
1065
1066 if (!res || errno != ENOSYS)
1067 return res;
1068#endif
1069
1070 /* even though we could play tricks with the flags, it's better to always
1071 * call fdatasync, as that matches the expectation of its users best */
1072 return fdatasync (fd);
1073}
1074
1075static int
1076eio__fallocate (int fd, int mode, off_t offset, size_t len)
1077{
1078#if HAVE_FALLOCATE
1079 return fallocate (fd, mode, offset, len);
1080#else
1081 errno = ENOSYS;
1082 return -1;
1083#endif
1084}
563 1085
564#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
1087# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 1089
1090static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1091eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 1092{
569 size_t todo = count; 1093 size_t todo = count;
570 dBUF; 1094 dBUF;
571 1095
572 while (todo > 0) 1096 while (todo > 0)
573 { 1097 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1098 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 1099
576 pread (fd, aio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
577 offset += len; 1101 offset += len;
578 todo -= len; 1102 todo -= len;
579 } 1103 }
580 1104
581 errno = 0; 1105 FUBd;
1106
1107 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1108 /* but not for e.g. EIO or eof, so we also never fail */
582 return count; 1109 return 0;
583} 1110}
584 1111
585#endif
586
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif 1112#endif
616 1113
617/* sendfile always needs emulation */ 1114/* sendfile always needs emulation */
1115static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1116eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 1117{
1118 eio_ssize_t written = 0;
620 ssize_t res; 1119 eio_ssize_t res;
621 1120
622 if (!count) 1121 if (!count)
623 return 0; 1122 return 0;
624 1123
1124 for (;;)
1125 {
1126#ifdef __APPLE__
1127# undef HAVE_SENDFILE /* broken, as everything on os x */
1128#endif
625#if HAVE_SENDFILE 1129#if HAVE_SENDFILE
626# if __linux 1130# if __linux
1131 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1132 res = sendfile (ofd, ifd, &soffset, count);
628 1133
629# elif __freebsd 1134# elif __FreeBSD__
630 /* 1135 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1136 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1137 * wasted on making it similar to other I/O functions.
633 */ 1138 */
634 {
635 off_t sbytes; 1139 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1140 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1141
638 if (res < 0 && sbytes) 1142 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1143 /* freebsd' sendfile will return 0 on success */
1144 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1145 /* not on e.g. EIO or EPIPE - sounds broken */
1146 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1147 res = sbytes;
641 } 1148 #endif
1149
1150 /* according to source inspection, this is correct, and useful behaviour */
1151 if (sbytes)
1152 res = sbytes;
1153
1154# elif defined (__APPLE__)
1155 off_t sbytes = count;
1156 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1157
1158 /* according to the manpage, sbytes is always valid */
1159 if (sbytes)
1160 res = sbytes;
642 1161
643# elif __hpux 1162# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1163 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1164
646# elif __solaris 1165# elif __solaris
647 {
648 struct sendfilevec vec; 1166 struct sendfilevec vec;
649 size_t sbytes; 1167 size_t sbytes;
650 1168
651 vec.sfv_fd = ifd; 1169 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1170 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1171 vec.sfv_off = offset;
654 vec.sfv_len = count; 1172 vec.sfv_len = count;
655 1173
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1174 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1175
658 if (res < 0 && sbytes) 1176 if (res < 0 && sbytes)
659 res = sbytes; 1177 res = sbytes;
660 }
661 1178
662# endif 1179# endif
1180
1181#elif defined (_WIN32) && 0
1182 /* does not work, just for documentation of what would need to be done */
1183 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1184 /* libeio guarantees that the file offset does not change, and windows */
1185 /* has no way to get an independent handle to the same file description */
1186 HANDLE h = TO_SOCKET (ifd);
1187 SetFilePointer (h, offset, 0, FILE_BEGIN);
1188 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1189
663#else 1190#else
664 res = -1; 1191 res = -1;
665 errno = ENOSYS; 1192 errno = ENOSYS;
666#endif 1193#endif
667 1194
1195 /* we assume sendfile can copy at least 128mb in one go */
1196 if (res <= 128 * 1024 * 1024)
1197 {
1198 if (res > 0)
1199 written += res;
1200
1201 if (written)
1202 return written;
1203
1204 break;
1205 }
1206 else
1207 {
1208 /* if we requested more, then probably the kernel was lazy */
1209 written += res;
1210 offset += res;
1211 count -= res;
1212
1213 if (!count)
1214 return written;
1215 }
1216 }
1217
668 if (res < 0 1218 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1219 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1220 /* BSDs */
1221#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1222 || errno == ENOTSUP
1223#endif
1224#ifdef EOPNOTSUPP /* windows */
1225 || errno == EOPNOTSUPP /* BSDs */
1226#endif
670#if __solaris 1227#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1228 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1229#endif
673 ) 1230 )
674 ) 1231 )
… …
678 1235
679 res = 0; 1236 res = 0;
680 1237
681 while (count) 1238 while (count)
682 { 1239 {
683 ssize_t cnt; 1240 eio_ssize_t cnt;
684 1241
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1242 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 1243
687 if (cnt <= 0) 1244 if (cnt <= 0)
688 { 1245 {
… …
700 1257
701 offset += cnt; 1258 offset += cnt;
702 res += cnt; 1259 res += cnt;
703 count -= cnt; 1260 count -= cnt;
704 } 1261 }
1262
1263 FUBd;
705 } 1264 }
706 1265
707 return res; 1266 return res;
708} 1267}
709 1268
710/* read a full directory */ 1269#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 1270# define eio_pagesize() PAGESIZE
1271#else
1272static intptr_t
1273eio_pagesize (void)
712{ 1274{
713 DIR *dirp; 1275 static intptr_t page;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp;
720 char *name, *names;
721 int memlen = 4096;
722 int memofs = 0;
723 int res = 0;
724 1276
725 X_LOCK (wrklock); 1277 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 1278 page = sysconf (_SC_PAGESIZE);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE;
729 req->ptr2 = names = malloc (memlen);
730 X_UNLOCK (wrklock);
731 1279
732 if (dirp && u && names) 1280 return page;
733 for (;;) 1281}
1282#endif
1283
1284static void
1285eio_page_align (void **addr, size_t *length)
1286{
1287 intptr_t mask = eio_pagesize () - 1;
1288
1289 /* round down addr */
1290 intptr_t adj = mask & (intptr_t)*addr;
1291
1292 *addr = (void *)((intptr_t)*addr - adj);
1293 *length += adj;
1294
1295 /* round up length */
1296 *length = (*length + mask) & ~mask;
1297}
1298
1299#if !_POSIX_MEMLOCK
1300# define eio__mlockall(a) EIO_ENOSYS ()
1301#else
1302
1303static int
1304eio__mlockall (int flags)
1305{
1306 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1307 extern int mallopt (int, int);
1308 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1309 #endif
1310
1311 if (EIO_MCL_CURRENT != MCL_CURRENT
1312 || EIO_MCL_FUTURE != MCL_FUTURE)
1313 {
1314 flags = 0
1315 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1316 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1317 }
1318
1319 return mlockall (flags);
1320}
1321#endif
1322
1323#if !_POSIX_MEMLOCK_RANGE
1324# define eio__mlock(a,b) EIO_ENOSYS ()
1325#else
1326
1327static int
1328eio__mlock (void *addr, size_t length)
1329{
1330 eio_page_align (&addr, &length);
1331
1332 return mlock (addr, length);
1333}
1334
1335#endif
1336
1337#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1338# define eio__msync(a,b,c) EIO_ENOSYS ()
1339#else
1340
1341static int
1342eio__msync (void *mem, size_t len, int flags)
1343{
1344 eio_page_align (&mem, &len);
1345
1346 if (EIO_MS_ASYNC != MS_SYNC
1347 || EIO_MS_INVALIDATE != MS_INVALIDATE
1348 || EIO_MS_SYNC != MS_SYNC)
1349 {
1350 flags = 0
1351 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1352 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1353 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1354 }
1355
1356 return msync (mem, len, flags);
1357}
1358
1359#endif
1360
1361static int
1362eio__mtouch (eio_req *req)
1363{
1364 void *mem = req->ptr2;
1365 size_t len = req->size;
1366 int flags = req->int1;
1367
1368 eio_page_align (&mem, &len);
1369
1370 {
1371 intptr_t addr = (intptr_t)mem;
1372 intptr_t end = addr + len;
1373 intptr_t page = eio_pagesize ();
1374
1375 if (addr < end)
1376 if (flags & EIO_MT_MODIFY) /* modify */
1377 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 else
1379 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1380 }
1381
1382 return 0;
1383}
1384
1385/*****************************************************************************/
1386/* requests implemented outside eio_execute, because they are so large */
1387
1388static void
1389eio__lseek (eio_req *req)
1390{
1391 /* this usually gets optimised away completely, or your compiler sucks, */
1392 /* or the whence constants really are not 0, 1, 2 */
1393 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1394 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1395 : req->int2 == EIO_SEEK_END ? SEEK_END
1396 : req->int2;
1397
1398 req->offs = lseek (req->int1, req->offs, whence);
1399 req->result = req->offs == (off_t)-1 ? -1 : 0;
1400}
1401
1402/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1403static int
1404eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1405{
1406 const char *rel = path;
1407 char *res;
1408 char *tmp1, *tmp2;
1409#if SYMLOOP_MAX > 32
1410 int symlinks = SYMLOOP_MAX;
1411#else
1412 int symlinks = 32;
1413#endif
1414
1415 errno = EINVAL;
1416 if (!rel)
1417 return -1;
1418
1419 errno = ENOENT;
1420 if (!*rel)
1421 return -1;
1422
1423 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1424 tmp1 = res + PATH_MAX;
1425 tmp2 = tmp1 + PATH_MAX;
1426
1427#if 0 /* disabled, the musl way to do things is just too racy */
1428#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1429 /* on linux we may be able to ask the kernel */
1430 {
1431 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1432
1433 if (fd >= 0)
734 { 1434 {
1435 sprintf (tmp1, "/proc/self/fd/%d", fd);
1436 req->result = readlink (tmp1, res, PATH_MAX);
1437 close (fd);
1438
1439 /* here we should probably stat the open file and the disk file, to make sure they still match */
1440
1441 if (req->result > 0)
1442 goto done;
1443 }
1444 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1445 return;
1446 }
1447#endif
1448#endif
1449
1450 if (*rel != '/')
1451 {
1452 int len;
1453
735 errno = 0; 1454 errno = ENOENT;
736 readdir_r (dirp, &u->d, &entp); 1455 if (wd == EIO_INVALID_WD)
1456 return -1;
1457
1458 if (wd == EIO_CWD)
1459 {
1460 if (!getcwd (res, PATH_MAX))
1461 return -1;
737 1462
738 if (!entp) 1463 len = strlen (res);
1464 }
1465 else
1466 memcpy (res, wd->str, len = wd->len);
1467
1468 if (res [1]) /* only use if not / */
1469 res += len;
1470 }
1471
1472 while (*rel)
1473 {
1474 eio_ssize_t len, linklen;
1475 const char *beg = rel;
1476
1477 while (*rel && *rel != '/')
1478 ++rel;
1479
1480 len = rel - beg;
1481
1482 if (!len) /* skip slashes */
1483 {
739 break; 1484 ++rel;
1485 continue;
1486 }
740 1487
741 name = entp->d_name; 1488 if (beg [0] == '.')
1489 {
1490 if (len == 1)
1491 continue; /* . - nop */
742 1492
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1493 if (beg [1] == '.' && len == 2)
1494 {
1495 /* .. - back up one component, if possible */
1496
1497 while (res != tmpbuf->ptr)
1498 if (*--res == '/')
1499 break;
1500
1501 continue;
1502 }
1503 }
1504
1505 errno = ENAMETOOLONG;
1506 if (res + 1 + len + 1 >= tmp1)
1507 return -1;
1508
1509 /* copy one component */
1510 *res = '/';
1511 memcpy (res + 1, beg, len);
1512
1513 /* zero-terminate, for readlink */
1514 res [len + 1] = 0;
1515
1516 /* now check if it's a symlink */
1517 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1518
1519 if (linklen < 0)
744 { 1520 {
745 int len = strlen (name) + 1; 1521 if (errno != EINVAL)
1522 return -1;
746 1523
747 res++; 1524 /* it's a normal directory. hopefully */
748
749 while (memofs + len > memlen)
750 {
751 memlen *= 2;
752 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen);
754 X_UNLOCK (wrklock);
755
756 if (!names)
757 break;
758 }
759
760 memcpy (names + memofs, name, len);
761 memofs += len; 1525 res += len + 1;
762 }
763 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769}
770
771/*****************************************************************************/
772
773X_THREAD_PROC (eio_proc)
774{
775 eio_req *req;
776 struct timespec ts;
777 worker *self = (worker *)thr_arg;
778
779 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781
782 for (;;)
783 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock);
787
788 for (;;)
789 {
790 self->req = req = reqq_shift (&req_queue);
791
792 if (req)
793 break;
794
795 ++idle;
796
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 {
799 if (idle > max_idle)
800 {
801 --idle;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 }
813
814 --idle;
815 }
816
817 --nready;
818
819 X_UNLOCK (reqlock);
820
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917
918 X_LOCK (reslock);
919
920 ++npending;
921
922 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb ();
924
925 self->req = 0;
926 worker_clear (self);
927
928 X_UNLOCK (reslock);
929 }
930
931quit:
932 X_LOCK (wrklock);
933 worker_free (self);
934 X_UNLOCK (wrklock);
935
936 return 0;
937}
938
939/*****************************************************************************/
940
941static void eio_atfork_prepare (void)
942{
943 X_LOCK (wrklock);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952}
953
954static void eio_atfork_parent (void)
955{
956#if !HAVE_READDIR_R
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965}
966
967static void eio_atfork_child (void)
968{
969 eio_req *prv;
970
971 while (prv = reqq_shift (&req_queue))
972 eio_destroy (prv);
973
974 while (prv = reqq_shift (&res_queue))
975 eio_destroy (prv);
976
977 while (wrk_first.next != &wrk_first)
978 {
979 worker *wrk = wrk_first.next;
980
981 if (wrk->req)
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 }
987
988 started = 0;
989 idle = 0;
990 nreqs = 0;
991 nready = 0;
992 npending = 0;
993
994 eio_atfork_parent ();
995}
996
997int eio_init (void (*want_poll)(void), void (*done_poll)(void))
998{
999 want_poll_cb = want_poll;
1000 done_poll_cb = done_poll;
1001
1002#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock);
1004 X_MUTEX_CHECK (reslock);
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 } 1526 }
1120 else 1527 else
1121 { 1528 {
1122 /* read: grow scalar as necessary */ 1529 /* yay, it was a symlink - build new path in tmp2 */
1123 svptr = SvGROW (data, len + dataoffset + 1); 1530 int rellen = strlen (rel);
1531
1532 errno = ENAMETOOLONG;
1533 if (linklen + 1 + rellen >= PATH_MAX)
1534 return -1;
1535
1536 errno = ELOOP;
1537 if (!--symlinks)
1538 return -1;
1539
1540 if (*tmp1 == '/')
1541 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1542
1543 /* we need to be careful, as rel might point into tmp2 already */
1544 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1545 tmp2 [linklen] = '/';
1546 memcpy (tmp2, tmp1, linklen);
1547
1548 rel = tmp2;
1124 } 1549 }
1550 }
1125 1551
1552 /* special case for the lone root path */
1553 if (res == tmpbuf->ptr)
1554 *res++ = '/';
1555
1556 return res - (char *)tmpbuf->ptr;
1557}
1558
1559static signed char
1560eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1561{
1562 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1563 : a->inode < b->inode ? -1
1564 : a->inode > b->inode ? 1
1565 : 0;
1566}
1567
1568#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1569
1570#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1571#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1572
1573static void
1574eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1575{
1576 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1577 unsigned char *bit = bits;
1578
1579 assert (CHAR_BIT == 8);
1580 assert (sizeof (eio_dirent) * 8 < 256);
1581 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1582 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1583
1584 if (size <= EIO_SORT_FAST)
1585 return;
1586
1587 /* first prepare an array of bits to test in our radix sort */
1588 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1589 /* inode_bits must contain all inodes ORed together */
1590 /* which is used to skip bits that are 0 everywhere, which is very common */
1591 {
1592 eio_ino_t endianness;
1593 int i, j;
1594
1595 /* we store the byte offset of byte n into byte n of "endianness" */
1596 for (i = 0; i < sizeof (eio_ino_t); ++i)
1597 ((unsigned char *)&endianness)[i] = i;
1598
1599 *bit++ = 0;
1600
1601 for (i = 0; i < sizeof (eio_ino_t); ++i)
1602 {
1603 /* shifting off the byte offsets out of "endianness" */
1604 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1605 endianness >>= 8;
1606
1607 for (j = 0; j < 8; ++j)
1608 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1609 *bit++ = offs + j;
1610 }
1611
1612 for (j = 0; j < 8; ++j)
1613 if (score_bits & (1 << j))
1614 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1615 }
1616
1617 /* now actually do the sorting (a variant of MSD radix sort) */
1618 {
1619 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1620 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1621 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1622 int stk_idx = 0;
1623
1624 base_stk [stk_idx] = dents;
1625 end_stk [stk_idx] = dents + size;
1626 bit_stk [stk_idx] = bit - 1;
1627
1628 do
1629 {
1630 base = base_stk [stk_idx];
1631 end = end_stk [stk_idx];
1632 bit = bit_stk [stk_idx];
1633
1634 for (;;)
1635 {
1636 unsigned char O = *bit >> 3;
1637 unsigned char M = 1 << (*bit & 7);
1638
1639 eio_dirent *a = base;
1640 eio_dirent *b = end;
1641
1642 if (b - a < EIO_SORT_CUTOFF)
1643 break;
1644
1645 /* now bit-partition the array on the bit */
1646 /* this ugly asymmetric loop seems to perform much better than typical */
1647 /* partition algos found in the literature */
1648 do
1649 if (!(((unsigned char *)a)[O] & M))
1650 ++a;
1651 else if (!(((unsigned char *)--b)[O] & M))
1652 {
1653 eio_dirent tmp = *a; *a = *b; *b = tmp;
1654 ++a;
1655 }
1656 while (b > a);
1657
1658 /* next bit, or stop, if no bits left in this path */
1659 if (!*--bit)
1660 break;
1661
1662 base_stk [stk_idx] = a;
1663 end_stk [stk_idx] = end;
1664 bit_stk [stk_idx] = bit;
1665 ++stk_idx;
1666
1667 end = a;
1668 }
1669 }
1670 while (stk_idx--);
1671 }
1672}
1673
1674static void
1675eio_dent_insertion_sort (eio_dirent *dents, int size)
1676{
1677 /* first move the smallest element to the front, to act as a sentinel */
1678 {
1679 int i;
1680 eio_dirent *min = dents;
1681
1682 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1683 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1684 if (EIO_DENT_CMP (dents [i], <, *min))
1685 min = &dents [i];
1686
1687 /* swap elements 0 and j (minimum) */
1688 {
1689 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1690 }
1691 }
1692
1693 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1694 {
1695 eio_dirent *i, *j;
1696
1697 for (i = dents + 1; i < dents + size; ++i)
1698 {
1699 eio_dirent value = *i;
1700
1701 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1702 j [1] = j [0];
1703
1704 j [1] = value;
1705 }
1706 }
1707}
1708
1709static void
1710eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1711{
1712 if (size <= 1)
1713 return; /* our insertion sort relies on size > 0 */
1714
1715 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1716 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1717 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1718
1719 /* use an insertion sort at the end, or for small arrays, */
1720 /* as insertion sort is more efficient for small partitions */
1721 eio_dent_insertion_sort (dents, size);
1722}
1723
1724/* read a full directory */
1725static void
1726eio__scandir (eio_req *req, etp_worker *self)
1727{
1728 char *name, *names;
1729 int namesalloc = 4096 - sizeof (void *) * 4;
1730 int namesoffs = 0;
1731 int flags = req->int1;
1732 eio_dirent *dents = 0;
1733 int dentalloc = 128;
1734 int dentoffs = 0;
1735 eio_ino_t inode_bits = 0;
1736#ifdef _WIN32
1737 HANDLE dirp;
1738 WIN32_FIND_DATA entp;
1739#else
1740 DIR *dirp;
1741 EIO_STRUCT_DIRENT *entp;
1742#endif
1743
1744 req->result = -1;
1745
1746 if (!(flags & EIO_READDIR_DENTS))
1747 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1748
1749#ifdef _WIN32
1750 {
1751 int len = strlen ((const char *)req->ptr1);
1752 char *path = malloc (MAX_PATH);
1753 const char *fmt;
1754 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1755
1756 if (!len)
1757 fmt = "./*";
1758 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1759 fmt = "%s*";
1760 else
1761 fmt = "%s/*";
1762
1763 _snprintf (path, MAX_PATH, fmt, reqpath);
1764 dirp = FindFirstFile (path, &entp);
1765 free (path);
1766
1767 if (dirp == INVALID_HANDLE_VALUE)
1768 {
1769 /* should steal _dosmaperr */
1770 switch (GetLastError ())
1771 {
1772 case ERROR_FILE_NOT_FOUND:
1773 req->result = 0;
1774 break;
1775
1776 case ERROR_INVALID_NAME:
1777 case ERROR_PATH_NOT_FOUND:
1778 case ERROR_NO_MORE_FILES:
1779 errno = ENOENT;
1780 break;
1781
1782 case ERROR_NOT_ENOUGH_MEMORY:
1783 errno = ENOMEM;
1784 break;
1785
1786 default:
1787 errno = EINVAL;
1788 break;
1789 }
1790
1791 return;
1792 }
1793 }
1794#else
1795 #if HAVE_AT
1796 if (req->wd)
1797 {
1798 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1799
1126 if (len < 0) 1800 if (fd < 0)
1127 croak ("length must not be negative"); 1801 return;
1128 1802
1803 dirp = fdopendir (fd);
1804
1805 if (!dirp)
1806 close (fd);
1807 }
1808 else
1809 dirp = opendir (req->ptr1);
1810 #else
1811 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1812 #endif
1813
1814 if (!dirp)
1815 return;
1816#endif
1817
1818 if (req->flags & EIO_FLAG_PTR1_FREE)
1819 free (req->ptr1);
1820
1821 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1822 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1823 req->ptr2 = names = malloc (namesalloc);
1824
1825 if (!names || (flags && !dents))
1826 return;
1827
1828 for (;;)
1829 {
1830 int done;
1831
1832#ifdef _WIN32
1833 done = !dirp;
1834#else
1835 errno = 0;
1836 entp = readdir (dirp);
1837 done = !entp;
1838#endif
1839
1840 if (done)
1129 { 1841 {
1130 dREQ; 1842#ifndef _WIN32
1843 int old_errno = errno;
1844 closedir (dirp);
1845 errno = old_errno;
1131 1846
1132 req->type = ix; 1847 if (errno)
1133 req->sv1 = newSVsv (fh); 1848 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1849#endif
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1850
1142 if (!SvREADONLY (data)) 1851 /* sort etc. */
1852 req->int1 = flags;
1853 req->result = dentoffs;
1854
1855 if (flags & EIO_READDIR_STAT_ORDER)
1856 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1857 else if (flags & EIO_READDIR_DIRS_FIRST)
1858 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1859 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1860 else
1861 {
1862 /* in this case, all is known, and we just put dirs first and sort them */
1863 eio_dirent *oth = dents + dentoffs;
1864 eio_dirent *dir = dents;
1865
1866 /* now partition dirs to the front, and non-dirs to the back */
1867 /* by walking from both sides and swapping if necessary */
1868 while (oth > dir)
1869 {
1870 if (dir->type == EIO_DT_DIR)
1871 ++dir;
1872 else if ((--oth)->type == EIO_DT_DIR)
1873 {
1874 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1875
1876 ++dir;
1877 }
1878 }
1879
1880 /* now sort the dirs only (dirs all have the same score) */
1881 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1882 }
1883
1884 break;
1885 }
1886
1887 /* now add the entry to our list(s) */
1888 name = D_NAME (entp);
1889
1890 /* skip . and .. entries */
1891 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1892 {
1893 int len = D_NAMLEN (entp) + 1;
1894
1895 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1896 {
1144 SvREADONLY_on (data); 1897 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1898 req->ptr2 = names = realloc (names, namesalloc);
1899
1900 if (!names)
1901 break;
1146 } 1902 }
1147 1903
1148 EIO_SEND; 1904 memcpy (names + namesoffs, name, len);
1905
1906 if (dents)
1907 {
1908 struct eio_dirent *ent;
1909
1910 if (ecb_expect_false (dentoffs == dentalloc))
1911 {
1912 dentalloc *= 2;
1913 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1914
1915 if (!dents)
1916 break;
1917 }
1918
1919 ent = dents + dentoffs;
1920
1921 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1922 ent->namelen = len - 1;
1923 ent->inode = D_INO (entp);
1924
1925 inode_bits |= ent->inode;
1926
1927 switch (D_TYPE (entp))
1928 {
1929 default:
1930 ent->type = EIO_DT_UNKNOWN;
1931 flags |= EIO_READDIR_FOUND_UNKNOWN;
1932 break;
1933
1934 #ifdef DT_FIFO
1935 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1936 #endif
1937 #ifdef DT_CHR
1938 case DT_CHR: ent->type = EIO_DT_CHR; break;
1939 #endif
1940 #ifdef DT_MPC
1941 case DT_MPC: ent->type = EIO_DT_MPC; break;
1942 #endif
1943 #ifdef DT_DIR
1944 case DT_DIR: ent->type = EIO_DT_DIR; break;
1945 #endif
1946 #ifdef DT_NAM
1947 case DT_NAM: ent->type = EIO_DT_NAM; break;
1948 #endif
1949 #ifdef DT_BLK
1950 case DT_BLK: ent->type = EIO_DT_BLK; break;
1951 #endif
1952 #ifdef DT_MPB
1953 case DT_MPB: ent->type = EIO_DT_MPB; break;
1954 #endif
1955 #ifdef DT_REG
1956 case DT_REG: ent->type = EIO_DT_REG; break;
1957 #endif
1958 #ifdef DT_NWK
1959 case DT_NWK: ent->type = EIO_DT_NWK; break;
1960 #endif
1961 #ifdef DT_CMP
1962 case DT_CMP: ent->type = EIO_DT_CMP; break;
1963 #endif
1964 #ifdef DT_LNK
1965 case DT_LNK: ent->type = EIO_DT_LNK; break;
1966 #endif
1967 #ifdef DT_SOCK
1968 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1969 #endif
1970 #ifdef DT_DOOR
1971 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1972 #endif
1973 #ifdef DT_WHT
1974 case DT_WHT: ent->type = EIO_DT_WHT; break;
1975 #endif
1976 }
1977
1978 ent->score = 7;
1979
1980 if (flags & EIO_READDIR_DIRS_FIRST)
1981 {
1982 if (ent->type == EIO_DT_UNKNOWN)
1983 {
1984 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1985 ent->score = 1;
1986 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1987 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1988 }
1989 else if (ent->type == EIO_DT_DIR)
1990 ent->score = 0;
1991 }
1992 }
1993
1994 namesoffs += len;
1995 ++dentoffs;
1149 } 1996 }
1997
1998 if (EIO_CANCELLED (req))
1999 {
2000 errno = ECANCELED;
2001 break;
2002 }
2003
2004#ifdef _WIN32
2005 if (!FindNextFile (dirp, &entp))
2006 {
2007 FindClose (dirp);
2008 dirp = 0;
2009 }
2010#endif
2011 }
2012}
2013
2014/*****************************************************************************/
2015/* working directory stuff */
2016/* various deficiencies in the posix 2008 api force us to */
2017/* keep the absolute path in string form at all times */
2018/* fuck yeah. */
2019
2020#if !HAVE_AT
2021
2022/* a bit like realpath, but usually faster because it doesn'T have to return */
2023/* an absolute or canonical path */
2024static const char *
2025wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2026{
2027 if (!wd || *path == '/')
2028 return path;
2029
2030 if (path [0] == '.' && !path [1])
2031 return wd->str;
2032
2033 {
2034 int l1 = wd->len;
2035 int l2 = strlen (path);
2036
2037 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2038
2039 memcpy (res, wd->str, l1);
2040 res [l1] = '/';
2041 memcpy (res + l1 + 1, path, l2 + 1);
2042
2043 return res;
2044 }
2045}
2046
2047#endif
2048
2049static eio_wd
2050eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2051{
2052 int fd;
2053 eio_wd res;
2054 int len = eio__realpath (tmpbuf, wd, path);
2055
2056 if (len < 0)
2057 return EIO_INVALID_WD;
2058
2059#if HAVE_AT
2060 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2061
2062 if (fd < 0)
2063 return EIO_INVALID_WD;
2064#endif
2065
2066 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2067
2068#if HAVE_AT
2069 res->fd = fd;
2070#endif
2071
2072 res->len = len;
2073 memcpy (res->str, tmpbuf->ptr, len);
2074 res->str [len] = 0;
2075
2076 return res;
2077}
2078
2079eio_wd
2080eio_wd_open_sync (eio_wd wd, const char *path)
2081{
2082 struct tmpbuf tmpbuf = { 0 };
2083 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2084 free (tmpbuf.ptr);
2085
2086 return wd;
1150} 2087}
1151 2088
1152void 2089void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 2090eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 2091{
1157 SV *data; 2092 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1158 dREQ; 2093 {
1159 2094 #if HAVE_AT
1160 data = newSV (NAME_MAX); 2095 close (wd->fd);
1161 SvPOK_on (data); 2096 #endif
1162 2097 free (wd);
1163 req->type = EIO_READLINK; 2098 }
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170} 2099}
2100
2101#if HAVE_AT
2102
2103/* they forgot these */
2104
2105static int
2106eio__truncateat (int dirfd, const char *path, off_t length)
2107{
2108 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2109 int res;
2110
2111 if (fd < 0)
2112 return fd;
2113
2114 res = ftruncate (fd, length);
2115 close (fd);
2116 return res;
2117}
2118
2119static int
2120eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2121{
2122 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2123 int res;
2124
2125 if (fd < 0)
2126 return fd;
2127
2128 res = fstatvfs (fd, buf);
2129 close (fd);
2130 return res;
2131
2132}
2133
2134#endif
2135
2136/*****************************************************************************/
2137
2138#define ALLOC(len) \
2139 if (!req->ptr2) \
2140 { \
2141 X_LOCK (wrklock); \
2142 req->flags |= EIO_FLAG_PTR2_FREE; \
2143 X_UNLOCK (wrklock); \
2144 req->ptr2 = malloc (len); \
2145 if (!req->ptr2) \
2146 { \
2147 errno = ENOMEM; \
2148 req->result = -1; \
2149 break; \
2150 } \
2151 }
2152
2153static void ecb_noinline ecb_cold
2154etp_proc_init (void)
2155{
2156#if HAVE_PRCTL_SET_NAME
2157 /* provide a more sensible "thread name" */
2158 char name[16 + 1];
2159 const int namelen = sizeof (name) - 1;
2160 int len;
2161
2162 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2163 name [namelen] = 0;
2164 len = strlen (name);
2165 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2166 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2167#endif
2168}
2169
2170X_THREAD_PROC (etp_proc)
2171{
2172 ETP_REQ *req;
2173 struct timespec ts;
2174 etp_worker *self = (etp_worker *)thr_arg;
2175
2176 etp_proc_init ();
2177
2178 /* try to distribute timeouts somewhat evenly */
2179 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
2180
2181 for (;;)
2182 {
2183 ts.tv_sec = 0;
2184
2185 X_LOCK (reqlock);
2186
2187 for (;;)
2188 {
2189 req = reqq_shift (&req_queue);
2190
2191 if (req)
2192 break;
2193
2194 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2195 {
2196 X_UNLOCK (reqlock);
2197 X_LOCK (wrklock);
2198 --started;
2199 X_UNLOCK (wrklock);
2200 goto quit;
2201 }
2202
2203 ++idle;
2204
2205 if (idle <= max_idle)
2206 /* we are allowed to idle, so do so without any timeout */
2207 X_COND_WAIT (reqwait, reqlock);
2208 else
2209 {
2210 /* initialise timeout once */
2211 if (!ts.tv_sec)
2212 ts.tv_sec = time (0) + idle_timeout;
2213
2214 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2215 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
2216 }
2217
2218 --idle;
2219 }
2220
2221 --nready;
2222
2223 X_UNLOCK (reqlock);
2224
2225 if (req->type < 0)
2226 goto quit;
2227
2228 ETP_EXECUTE (self, req);
2229
2230 X_LOCK (reslock);
2231
2232 ++npending;
2233
2234 if (!reqq_push (&res_queue, req) && want_poll_cb)
2235 want_poll_cb ();
2236
2237 etp_worker_clear (self);
2238
2239 X_UNLOCK (reslock);
2240 }
2241
2242quit:
2243 free (req);
2244
2245 X_LOCK (wrklock);
2246 etp_worker_free (self);
2247 X_UNLOCK (wrklock);
2248
2249 return 0;
2250}
2251
2252/*****************************************************************************/
2253
2254int ecb_cold
2255eio_init (void (*want_poll)(void), void (*done_poll)(void))
2256{
2257#if !HAVE_PREADWRITE
2258 X_MUTEX_CREATE (preadwritelock);
2259#endif
2260
2261 return etp_init (want_poll, done_poll);
2262}
2263
2264ecb_inline void
2265eio_api_destroy (eio_req *req)
2266{
2267 free (req);
2268}
2269
2270#define REQ(rtype) \
2271 eio_req *req; \
2272 \
2273 req = (eio_req *)calloc (1, sizeof *req); \
2274 if (!req) \
2275 return 0; \
2276 \
2277 req->type = rtype; \
2278 req->pri = pri; \
2279 req->finish = cb; \
2280 req->data = data; \
2281 req->destroy = eio_api_destroy;
2282
2283#define SEND eio_submit (req); return req
2284
2285#define PATH \
2286 req->flags |= EIO_FLAG_PTR1_FREE; \
2287 req->ptr1 = strdup (path); \
2288 if (!req->ptr1) \
2289 { \
2290 eio_api_destroy (req); \
2291 return 0; \
2292 }
2293
2294static void
2295eio_execute (etp_worker *self, eio_req *req)
2296{
2297#if HAVE_AT
2298 int dirfd;
2299#else
2300 const char *path;
2301#endif
2302
2303 if (ecb_expect_false (EIO_CANCELLED (req)))
2304 {
2305 req->result = -1;
2306 req->errorno = ECANCELED;
2307 return;
2308 }
2309
2310 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2311 {
2312 req->result = -1;
2313 req->errorno = ENOENT;
2314 return;
2315 }
2316
2317 if (req->type >= EIO_OPEN)
2318 {
2319 #if HAVE_AT
2320 dirfd = WD2FD (req->wd);
2321 #else
2322 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2323 #endif
2324 }
2325
2326 switch (req->type)
2327 {
2328 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2329 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2330 break;
2331 case EIO_WD_CLOSE: req->result = 0;
2332 eio_wd_close_sync (req->wd); break;
2333
2334 case EIO_SEEK: eio__lseek (req); break;
2335 case EIO_READ: ALLOC (req->size);
2336 req->result = req->offs >= 0
2337 ? pread (req->int1, req->ptr2, req->size, req->offs)
2338 : read (req->int1, req->ptr2, req->size); break;
2339 case EIO_WRITE: req->result = req->offs >= 0
2340 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2341 : write (req->int1, req->ptr2, req->size); break;
2342
2343 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2344 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2345
2346#if HAVE_AT
2347
2348 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2349 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2350 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2351 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2352 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2353 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2354 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2355 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2356
2357 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2358 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2359 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2360 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2361 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2362 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2363 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2364 case EIO_READLINK: ALLOC (PATH_MAX);
2365 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2366 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2367 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2368 case EIO_UTIME:
2369 case EIO_FUTIME:
2370 {
2371 struct timespec ts[2];
2372 struct timespec *times;
2373
2374 if (req->nv1 != -1. || req->nv2 != -1.)
2375 {
2376 ts[0].tv_sec = req->nv1;
2377 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2378 ts[1].tv_sec = req->nv2;
2379 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2380
2381 times = ts;
2382 }
2383 else
2384 times = 0;
2385
2386 req->result = req->type == EIO_FUTIME
2387 ? futimens (req->int1, times)
2388 : utimensat (dirfd, req->ptr1, times, 0);
2389 }
2390 break;
2391
2392#else
2393
2394 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2395 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2396 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2397 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2398 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2399 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2400 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2401 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2402
2403 case EIO_UNLINK: req->result = unlink (path ); break;
2404 case EIO_RMDIR: req->result = rmdir (path ); break;
2405 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2406 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2407 case EIO_LINK: req->result = link (path , req->ptr2); break;
2408 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2409 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2410 case EIO_READLINK: ALLOC (PATH_MAX);
2411 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2412 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2413 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2414
2415 case EIO_UTIME:
2416 case EIO_FUTIME:
2417 {
2418 struct timeval tv[2];
2419 struct timeval *times;
2420
2421 if (req->nv1 != -1. || req->nv2 != -1.)
2422 {
2423 tv[0].tv_sec = req->nv1;
2424 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2425 tv[1].tv_sec = req->nv2;
2426 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2427
2428 times = tv;
2429 }
2430 else
2431 times = 0;
2432
2433 req->result = req->type == EIO_FUTIME
2434 ? futimes (req->int1, times)
2435 : utimes (req->ptr1, times);
2436 }
2437 break;
2438
2439#endif
2440
2441 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2442 {
2443 ALLOC (req->result);
2444 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2445 }
2446 break;
2447
2448 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2449 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2450
2451 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2452 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2453
2454 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2455 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2456 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2457
2458 case EIO_CLOSE: req->result = close (req->int1); break;
2459 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2460 case EIO_SYNC: req->result = 0; sync (); break;
2461 case EIO_FSYNC: req->result = fsync (req->int1); break;
2462 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2463 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2464 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2465 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2466 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2467 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2468 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2469 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2470
2471 case EIO_READDIR: eio__scandir (req, self); break;
2472
2473 case EIO_BUSY:
2474#ifdef _WIN32
2475 Sleep (req->nv1 * 1e3);
2476#else
2477 {
2478 struct timeval tv;
2479
2480 tv.tv_sec = req->nv1;
2481 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2482
2483 req->result = select (0, 0, 0, 0, &tv);
2484 }
2485#endif
2486 break;
2487
2488 case EIO_GROUP:
2489 abort (); /* handled in eio_request */
2490
2491 case EIO_NOP:
2492 req->result = 0;
2493 break;
2494
2495 case EIO_CUSTOM:
2496 req->feed (req);
2497 break;
2498
2499 default:
2500 errno = ENOSYS;
2501 req->result = -1;
2502 break;
2503 }
2504
2505 req->errorno = errno;
2506}
2507
2508#ifndef EIO_NO_WRAPPERS
2509
2510eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2511{
2512 REQ (EIO_WD_OPEN); PATH; SEND;
2513}
2514
2515eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2516{
2517 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2518}
2519
2520eio_req *eio_nop (int pri, eio_cb cb, void *data)
2521{
2522 REQ (EIO_NOP); SEND;
2523}
2524
2525eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2526{
2527 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2528}
2529
2530eio_req *eio_sync (int pri, eio_cb cb, void *data)
2531{
2532 REQ (EIO_SYNC); SEND;
2533}
2534
2535eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2536{
2537 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2538}
2539
2540eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2541{
2542 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2543}
2544
2545eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2546{
2547 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2548}
2549
2550eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2551{
2552 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2553}
2554
2555eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2556{
2557 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2558}
2559
2560eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2561{
2562 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2563}
2564
2565eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2566{
2567 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2568}
2569
2570eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2571{
2572 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2573}
2574
2575eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2576{
2577 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2578}
2579
2580eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2581{
2582 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2583}
2584
2585eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2586{
2587 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2588}
2589
2590eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2591{
2592 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2593}
2594
2595eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2596{
2597 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2598}
2599
2600eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2601{
2602 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2603}
2604
2605eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2606{
2607 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2608}
2609
2610eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2611{
2612 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2613}
2614
2615eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2616{
2617 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2618}
2619
2620eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2621{
2622 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2623}
2624
2625eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2626{
2627 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2628}
2629
2630eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2631{
2632 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2633}
2634
2635eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2636{
2637 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2638}
2639
2640eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2641{
2642 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2643}
2644
2645eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2646{
2647 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2648}
2649
2650eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2651{
2652 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2653}
2654
2655eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2656{
2657 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2658}
2659
2660eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2661{
2662 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2663}
2664
2665eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2666{
2667 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2668}
2669
2670eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2671{
2672 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2673}
2674
2675static eio_req *
2676eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2677{
2678 REQ (type); PATH; SEND;
2679}
2680
2681eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2682{
2683 return eio__1path (EIO_READLINK, path, pri, cb, data);
2684}
2685
2686eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2687{
2688 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2689}
2690
2691eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2692{
2693 return eio__1path (EIO_STAT, path, pri, cb, data);
2694}
2695
2696eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2697{
2698 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2699}
2700
2701eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2702{
2703 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2704}
2705
2706eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2707{
2708 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2709}
2710
2711eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2712{
2713 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2714}
2715
2716eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2717{
2718 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2719}
2720
2721eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2722{
2723 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2724}
2725
2726static eio_req *
2727eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2728{
2729 REQ (type); PATH;
2730
2731 req->flags |= EIO_FLAG_PTR2_FREE;
2732 req->ptr2 = strdup (new_path);
2733 if (!req->ptr2)
2734 {
2735 eio_api_destroy (req);
2736 return 0;
2737 }
2738
2739 SEND;
2740}
2741
2742eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2743{
2744 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2745}
2746
2747eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2748{
2749 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2750}
2751
2752eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2753{
2754 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2755}
2756
2757eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2758{
2759 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2760}
2761
2762#endif
2763
2764eio_req *eio_grp (eio_cb cb, void *data)
2765{
2766 const int pri = EIO_PRI_MAX;
2767
2768 REQ (EIO_GROUP); SEND;
2769}
2770
2771#undef REQ
2772#undef PATH
2773#undef SEND
2774
2775/*****************************************************************************/
2776/* grp functions */
1171 2777
1172void 2778void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2779eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2780{
1421 grp->int2 = limit; 2781 grp->int2 = limit;
1422 grp->feed = feed; 2782 grp->feed = feed;
1423 2783
1424 grp_try_feed (grp); 2784 grp_try_feed (grp);
1425} 2785}
1426 2786
2787void
1427void eio_grp_limit (eio_req *grp, int limit) 2788eio_grp_limit (eio_req *grp, int limit)
1428{ 2789{
1429 grp->int2 = limit; 2790 grp->int2 = limit;
1430 2791
1431 grp_try_feed (grp); 2792 grp_try_feed (grp);
1432} 2793}
1433 2794
2795void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2796eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2797{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2799
2800 grp->flags |= EIO_FLAG_GROUPADD;
1437 2801
1438 ++grp->size; 2802 ++grp->size;
1439 req->grp = grp; 2803 req->grp = grp;
1440 2804
1441 req->grp_prev = 0; 2805 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2809 grp->grp_first->grp_prev = req;
1446 2810
1447 grp->grp_first = req; 2811 grp->grp_first = req;
1448} 2812}
1449 2813
2814/*****************************************************************************/
2815/* misc garbage */
1450 2816
2817eio_ssize_t
2818eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2819{
2820 return eio__sendfile (ofd, ifd, offset, count);
2821}
2822

Diff Legend

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