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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.115 by root, Tue Mar 27 18:54:45 2012 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
… …
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
15#ifndef EIO_FINISH 82#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 84#endif
18 85
19#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
22 89
23#ifndef EIO_FEED 90#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 92#endif
26 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
27#ifdef _WIN32 105#ifdef _WIN32
28 106
29 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 196
31#else 197#else
32 198
33# include "config.h"
34# include <sys/time.h> 199 #include <sys/time.h>
35# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
36# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
37# include <utime.h> 237# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 238#endif
44 239
45#endif 240#if HAVE_SYS_SYSCALL_H
46 241# include <sys/syscall.h>
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif 242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
246#endif
50 247
51#if HAVE_SENDFILE 248#if HAVE_SENDFILE
52# if __linux 249# if __linux
53# include <sys/sendfile.h> 250# include <sys/sendfile.h>
54# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 252# include <sys/socket.h>
56# include <sys/uio.h> 253# include <sys/uio.h>
57# elif __hpux 254# elif __hpux
58# include <sys/socket.h> 255# include <sys/socket.h>
59# elif __solaris /* not yet */ 256# elif __solaris
60# include <sys/sendfile.h> 257# include <sys/sendfile.h>
61# else 258# else
62# error sendfile support requested but not available 259# error sendfile support requested but not available
63# endif 260# endif
64#endif 261#endif
65 262
66/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
68 272
69/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 274#ifndef NAME_MAX
71# define NAME_MAX 4096 275# define NAME_MAX 4096
72#endif 276#endif
73 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
74/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
76 285
77#define dBUF \ 286#define dBUF \
78 char *eio_buf; \
79 X_LOCK (wrklock); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 288 errno = ENOMEM; \
82 if (!eio_buf) \ 289 if (!eio_buf) \
83 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
84 294
85#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
340
341#define ETP_PRI_MIN EIO_PRI_MIN
342#define ETP_PRI_MAX EIO_PRI_MAX
343
344struct etp_worker;
345
346#define ETP_REQ eio_req
347#define ETP_DESTROY(req) eio_destroy (req)
348static int eio_finish (eio_req *req);
349#define ETP_FINISH(req) eio_finish (req)
350static void eio_execute (struct etp_worker *self, eio_req *req);
351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
352
353/*****************************************************************************/
354
355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
356
357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
359tvdiff (struct timeval *tv1, struct timeval *tv2)
360{
361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
363}
364
365static unsigned int started, idle, wanted = 4;
86 366
87static void (*want_poll_cb) (void); 367static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
89 369
90static unsigned int max_poll_time = 0; 370static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 371static unsigned int max_poll_reqs; /* reslock */
92 372
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 373static unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 374static unsigned int nready; /* reqlock */
95{ 375static unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 378
100static unsigned int started, idle, wanted; 379static xmutex_t wrklock;
380static xmutex_t reslock;
381static xmutex_t reqlock;
382static xcond_t reqwait;
383
384#if !HAVE_PREADWRITE
385/*
386 * make our pread/pwrite emulation safe against themselves, but not against
387 * normal read/write by using a mutex. slows down execution a lot,
388 * but that's your problem, not mine.
389 */
390static xmutex_t preadwritelock;
391#endif
392
393typedef struct etp_worker
394{
395 struct tmpbuf tmpbuf;
396
397 /* locked by wrklock */
398 struct etp_worker *prev, *next;
399
400 xthread_t tid;
401
402#ifdef ETP_WORKER_COMMON
403 ETP_WORKER_COMMON
404#endif
405} etp_worker;
406
407static etp_worker wrk_first; /* NOT etp */
408
409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 411
102/* worker threads management */ 412/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 413
105typedef struct worker { 414static void
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 415etp_worker_clear (etp_worker *wrk)
120{ 416{
121 if (wrk->dirp) 417}
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126 418
127 if (wrk->dbuf) 419static void ecb_cold
128 { 420etp_worker_free (etp_worker *wrk)
421{
129 free (wrk->dbuf); 422 free (wrk->tmpbuf.ptr);
130 wrk->dbuf = 0;
131 }
132}
133 423
134static void worker_free (worker *wrk)
135{
136 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
138 426
139 free (wrk); 427 free (wrk);
140} 428}
141 429
142static volatile unsigned int nreqs, nready, npending; 430static unsigned int
143static volatile unsigned int max_idle = 4; 431etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 432{
433 int retval;
434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
435 retval = nreqs;
436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 437 return retval;
152} 438}
153 439
154unsigned int eio_nready (void) 440static unsigned int
441etp_nready (void)
155{ 442{
156 unsigned int retval; 443 unsigned int retval;
157 444
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 446 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 448
162 return retval; 449 return retval;
163} 450}
164 451
165unsigned int eio_npending (void) 452static unsigned int
453etp_npending (void)
166{ 454{
167 unsigned int retval; 455 unsigned int retval;
168 456
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 458 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 460
173 return retval; 461 return retval;
174} 462}
175 463
176unsigned int eio_nthreads (void) 464static unsigned int
465etp_nthreads (void)
177{ 466{
178 unsigned int retval; 467 unsigned int retval;
179 468
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 470 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 477 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 478 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 479 * per shift, the most expensive operation.
191 */ 480 */
192typedef struct { 481typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 482 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 483 int size;
195} reqq; 484} etp_reqq;
196 485
197static reqq req_queue; 486static etp_reqq req_queue;
198static reqq res_queue; 487static etp_reqq res_queue;
199 488
200static int reqq_push (reqq *q, eio_req *req) 489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
501reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 502{
202 int pri = req->pri; 503 int pri = req->pri;
203 req->next = 0; 504 req->next = 0;
204 505
205 if (q->qe[pri]) 506 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
212 513
213 return q->size++; 514 return q->size++;
214} 515}
215 516
216static eio_req *reqq_shift (reqq *q) 517static ETP_REQ * ecb_noinline
518reqq_shift (etp_reqq *q)
217{ 519{
218 int pri; 520 int pri;
219 521
220 if (!q->size) 522 if (!q->size)
221 return 0; 523 return 0;
222 524
223 --q->size; 525 --q->size;
224 526
225 for (pri = EIO_NUM_PRI; pri--; ) 527 for (pri = ETP_NUM_PRI; pri--; )
226 { 528 {
227 eio_req *req = q->qs[pri]; 529 eio_req *req = q->qs[pri];
228 530
229 if (req) 531 if (req)
230 { 532 {
… …
236 } 538 }
237 539
238 abort (); 540 abort ();
239} 541}
240 542
241static void grp_try_feed (eio_req *grp) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
242{ 545{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 546 X_MUTEX_CREATE (wrklock);
244 { 547 X_MUTEX_CREATE (reslock);
245 int old_len = grp->size; 548 X_MUTEX_CREATE (reqlock);
549 X_COND_CREATE (reqwait);
246 550
247 EIO_FEED (grp); 551 reqq_init (&req_queue);
552 reqq_init (&res_queue);
248 553
249 /* stop if no progress has been made */ 554 wrk_first.next =
250 if (old_len == grp->size) 555 wrk_first.prev = &wrk_first;
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256}
257 556
258static int eio_finish (eio_req *req); 557 started = 0;
558 idle = 0;
559 nreqs = 0;
560 nready = 0;
561 npending = 0;
259 562
260static int grp_dec (eio_req *grp) 563 want_poll_cb = want_poll;
261{ 564 done_poll_cb = done_poll;
262 --grp->size;
263 565
264 /* call feeder, if applicable */
265 grp_try_feed (grp);
266
267 /* finish, if done */
268 if (!grp->size && grp->int1)
269 return eio_finish (grp);
270 else
271 return 0; 566 return 0;
272} 567}
273 568
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 569X_THREAD_PROC (etp_proc);
325 570
326static void start_thread (void) 571static void ecb_cold
572etp_start_thread (void)
327{ 573{
328 worker *wrk = calloc (1, sizeof (worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 575
330 /*TODO*/ 576 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 577 assert (("unable to allocate worker thread data", wrk));
332 578
333 X_LOCK (wrklock); 579 X_LOCK (wrklock);
334 580
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 581 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 582 {
337 wrk->prev = &wrk_first; 583 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 584 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 585 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 586 wrk_first.next = wrk;
… …
344 free (wrk); 590 free (wrk);
345 591
346 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
347} 593}
348 594
595static void
349static void maybe_start_thread (void) 596etp_maybe_start_thread (void)
350{ 597{
351 if (eio_nthreads () >= wanted) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 599 return;
353 600
354 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
355 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 ()))
356 return; 603 return;
357 604
358 start_thread (); 605 etp_start_thread ();
359} 606}
360 607
361void eio_submit (eio_req *req) 608static void ecb_cold
609etp_end_thread (void)
362{ 610{
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void)
375{
376 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
377 612
378 req->type = EIO_QUIT; 613 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 615
381 X_LOCK (reqlock); 616 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 621 X_LOCK (wrklock);
387 --started; 622 --started;
388 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
389} 624}
390 625
391void eio_set_max_poll_time (double nseconds) 626static int
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 627etp_poll (void)
428{ 628{
429 int maxreqs = max_poll_reqs; 629 unsigned int maxreqs;
630 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
431 eio_req *req;
432 632
633 X_LOCK (reslock);
634 maxreqs = max_poll_reqs;
635 maxtime = max_poll_time;
636 X_UNLOCK (reslock);
637
433 if (max_poll_time) 638 if (maxtime)
434 gettimeofday (&tv_start, 0); 639 gettimeofday (&tv_start, 0);
435 640
436 for (;;) 641 for (;;)
437 { 642 {
643 ETP_REQ *req;
644
438 maybe_start_thread (); 645 etp_maybe_start_thread ();
439 646
440 X_LOCK (reslock); 647 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 648 req = reqq_shift (&res_queue);
442 649
443 if (req) 650 if (req)
… …
451 X_UNLOCK (reslock); 658 X_UNLOCK (reslock);
452 659
453 if (!req) 660 if (!req)
454 return 0; 661 return 0;
455 662
663 X_LOCK (reqlock);
456 --nreqs; 664 --nreqs;
665 X_UNLOCK (reqlock);
457 666
458 if (req->type == EIO_GROUP && req->size) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 668 {
460 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
461 continue; 670 continue;
462 } 671 }
463 else 672 else
464 { 673 {
465 int res = eio_finish (req); 674 int res = ETP_FINISH (req);
466 if (res) 675 if (ecb_expect_false (res))
467 return res; 676 return res;
468 } 677 }
469 678
470 if (maxreqs && !--maxreqs) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 680 break;
472 681
473 if (max_poll_time) 682 if (maxtime)
474 { 683 {
475 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
476 685
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 686 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 687 break;
479 } 688 }
480 } 689 }
481 690
482 errno = EAGAIN; 691 errno = EAGAIN;
483 return -1; 692 return -1;
484} 693}
485 694
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
486/*****************************************************************************/ 945/*****************************************************************************/
487/* work around various missing functions */ 946/* work around various missing functions */
488 947
489#if !HAVE_PREADWRITE 948#if !HAVE_PREADWRITE
949# undef pread
950# undef pwrite
490# define pread aio_pread 951# define pread eio__pread
491# define pwrite aio_pwrite 952# define pwrite eio__pwrite
492 953
493/* 954static eio_ssize_t
494 * make our pread/pwrite safe against themselves, but not against 955eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 956{
502 ssize_t res; 957 eio_ssize_t res;
503 off_t ooffset; 958 off_t ooffset;
504 959
505 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
506 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
507 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
510 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
511 966
512 return res; 967 return res;
513} 968}
514 969
515static 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)
516{ 972{
517 ssize_t res; 973 eio_ssize_t res;
518 off_t ooffset; 974 off_t ooffset;
519 975
520 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 979 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 980 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 981 X_UNLOCK (preadwritelock);
526 982
527 return res; 983 return res;
528} 984}
529#endif 985#endif
530 986
531#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
532 988
989# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 991
992static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
537{ 994{
538 if (times) 995 if (times)
539 { 996 {
540 struct utimbuf buf; 997 struct utimbuf buf;
541 998
… …
546 } 1003 }
547 else 1004 else
548 return utime (filename, 0); 1005 return utime (filename, 0);
549} 1006}
550 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
552{ 1017{
553 errno = ENOSYS; 1018 errno = ENOSYS;
554 return -1; 1019 return -1;
555} 1020}
556 1021
557#endif 1022#endif
558 1023
559#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
1025# undef fdatasync
560# define fdatasync fsync 1026# define fdatasync(fd) fsync (fd)
561#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}
562 1085
563#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
1087# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 1089
1090static eio_ssize_t
566static 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)
567{ 1092{
568 size_t todo = count; 1093 size_t todo = count;
569 dBUF; 1094 dBUF;
570 1095
571 while (todo > 0) 1096 while (todo > 0)
572 { 1097 {
573 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1098 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
574
575 pread (fd, eio_buf, len, offset);
576 offset += len; 1099 offset += len;
577 todo -= len; 1100 todo -= len;
578 } 1101 }
579 1102
580 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
581 return count; 1107 return 0;
582} 1108}
583 1109
584#endif
585
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif 1110#endif
615 1111
616/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
1113static eio_ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
618{ 1115{
1116 eio_ssize_t written = 0;
619 ssize_t res; 1117 eio_ssize_t res;
620 1118
621 if (!count) 1119 if (!count)
622 return 0; 1120 return 0;
623 1121
1122 for (;;)
1123 {
1124#ifdef __APPLE__
1125# undef HAVE_SENDFILE /* broken, as everything on os x */
1126#endif
624#if HAVE_SENDFILE 1127#if HAVE_SENDFILE
625# if __linux 1128# if __linux
1129 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1130 res = sendfile (ofd, ifd, &soffset, count);
627 1131
628# elif __freebsd 1132# elif __FreeBSD__
629 /* 1133 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 1134 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1135 * wasted on making it similar to other I/O functions.
632 */ 1136 */
633 {
634 off_t sbytes; 1137 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1138 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1139
637 if (res < 0 && sbytes) 1140 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1141 /* freebsd' sendfile will return 0 on success */
1142 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1143 /* not on e.g. EIO or EPIPE - sounds broken */
1144 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1145 res = sbytes;
640 } 1146 #endif
1147
1148 /* according to source inspection, this is correct, and useful behaviour */
1149 if (sbytes)
1150 res = sbytes;
1151
1152# elif defined (__APPLE__)
1153 off_t sbytes = count;
1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1155
1156 /* according to the manpage, sbytes is always valid */
1157 if (sbytes)
1158 res = sbytes;
641 1159
642# elif __hpux 1160# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1161 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1162
645# elif __solaris 1163# elif __solaris
646 {
647 struct sendfilevec vec; 1164 struct sendfilevec vec;
648 size_t sbytes; 1165 size_t sbytes;
649 1166
650 vec.sfv_fd = ifd; 1167 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1168 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1169 vec.sfv_off = offset;
653 vec.sfv_len = count; 1170 vec.sfv_len = count;
654 1171
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1172 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1173
657 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
658 res = sbytes; 1175 res = sbytes;
659 }
660 1176
661# endif 1177# endif
1178
1179#elif defined (_WIN32) && 0
1180 /* does not work, just for documentation of what would need to be done */
1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1182 /* libeio guarantees that the file offset does not change, and windows */
1183 /* has no way to get an independent handle to the same file description */
1184 HANDLE h = TO_SOCKET (ifd);
1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1187
662#else 1188#else
663 res = -1; 1189 res = -1;
664 errno = ENOSYS; 1190 errno = ENOSYS;
665#endif 1191#endif
666 1192
1193 /* we assume sendfile can copy at least 128mb in one go */
1194 if (res <= 128 * 1024 * 1024)
1195 {
1196 if (res > 0)
1197 written += res;
1198
1199 if (written)
1200 return written;
1201
1202 break;
1203 }
1204 else
1205 {
1206 /* if we requested more, then probably the kernel was lazy */
1207 written += res;
1208 offset += res;
1209 count -= res;
1210
1211 if (!count)
1212 return written;
1213 }
1214 }
1215
667 if (res < 0 1216 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1217 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1218 /* BSDs */
1219#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1220 || errno == ENOTSUP
1221#endif
1222#ifdef EOPNOTSUPP /* windows */
1223 || errno == EOPNOTSUPP /* BSDs */
1224#endif
669#if __solaris 1225#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1226 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1227#endif
672 ) 1228 )
673 ) 1229 )
… …
677 1233
678 res = 0; 1234 res = 0;
679 1235
680 while (count) 1236 while (count)
681 { 1237 {
682 ssize_t cnt; 1238 eio_ssize_t cnt;
683 1239
684 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1240 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
685 1241
686 if (cnt <= 0) 1242 if (cnt <= 0)
687 { 1243 {
… …
699 1255
700 offset += cnt; 1256 offset += cnt;
701 res += cnt; 1257 res += cnt;
702 count -= cnt; 1258 count -= cnt;
703 } 1259 }
1260
1261 FUBd;
704 } 1262 }
705 1263
706 return res; 1264 return res;
707} 1265}
708 1266
709/* read a full directory */ 1267#ifdef PAGESIZE
710static void scandir_ (eio_req *req, worker *self) 1268# define eio_pagesize() PAGESIZE
1269#else
1270static intptr_t
1271eio_pagesize (void)
711{ 1272{
712 DIR *dirp; 1273 static intptr_t page;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp;
719 char *name, *names;
720 int memlen = 4096;
721 int memofs = 0;
722 int res = 0;
723 1274
724 X_LOCK (wrklock); 1275 if (!page)
725 self->dirp = dirp = opendir (req->ptr1); 1276 page = sysconf (_SC_PAGESIZE);
726 self->dbuf = u = malloc (sizeof (*u));
727 req->flags |= EIO_FLAG_PTR2_FREE;
728 req->ptr2 = names = malloc (memlen);
729 X_UNLOCK (wrklock);
730 1277
731 if (dirp && u && names) 1278 return page;
732 for (;;) 1279}
1280#endif
1281
1282static void
1283eio_page_align (void **addr, size_t *length)
1284{
1285 intptr_t mask = eio_pagesize () - 1;
1286
1287 /* round down addr */
1288 intptr_t adj = mask & (intptr_t)*addr;
1289
1290 *addr = (void *)((intptr_t)*addr - adj);
1291 *length += adj;
1292
1293 /* round up length */
1294 *length = (*length + mask) & ~mask;
1295}
1296
1297#if !_POSIX_MEMLOCK
1298# define eio__mlockall(a) EIO_ENOSYS ()
1299#else
1300
1301static int
1302eio__mlockall (int flags)
1303{
1304 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1305 extern int mallopt (int, int);
1306 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1307 #endif
1308
1309 if (EIO_MCL_CURRENT != MCL_CURRENT
1310 || EIO_MCL_FUTURE != MCL_FUTURE)
1311 {
1312 flags = 0
1313 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1314 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1315 }
1316
1317 return mlockall (flags);
1318}
1319#endif
1320
1321#if !_POSIX_MEMLOCK_RANGE
1322# define eio__mlock(a,b) EIO_ENOSYS ()
1323#else
1324
1325static int
1326eio__mlock (void *addr, size_t length)
1327{
1328 eio_page_align (&addr, &length);
1329
1330 return mlock (addr, length);
1331}
1332
1333#endif
1334
1335#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1336# define eio__msync(a,b,c) EIO_ENOSYS ()
1337#else
1338
1339static int
1340eio__msync (void *mem, size_t len, int flags)
1341{
1342 eio_page_align (&mem, &len);
1343
1344 if (EIO_MS_ASYNC != MS_SYNC
1345 || EIO_MS_INVALIDATE != MS_INVALIDATE
1346 || EIO_MS_SYNC != MS_SYNC)
1347 {
1348 flags = 0
1349 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1350 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1351 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1352 }
1353
1354 return msync (mem, len, flags);
1355}
1356
1357#endif
1358
1359static int
1360eio__mtouch (eio_req *req)
1361{
1362 void *mem = req->ptr2;
1363 size_t len = req->size;
1364 int flags = req->int1;
1365
1366 eio_page_align (&mem, &len);
1367
1368 {
1369 intptr_t addr = (intptr_t)mem;
1370 intptr_t end = addr + len;
1371 intptr_t page = eio_pagesize ();
1372
1373 if (addr < end)
1374 if (flags & EIO_MT_MODIFY) /* modify */
1375 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1376 else
1377 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 }
1379
1380 return 0;
1381}
1382
1383/*****************************************************************************/
1384/* requests implemented outside eio_execute, because they are so large */
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 errno = EINVAL;
1400 if (!rel)
1401 return -1;
1402
1403 errno = ENOENT;
1404 if (!*rel)
1405 return -1;
1406
1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1408 tmp1 = res + PATH_MAX;
1409 tmp2 = tmp1 + PATH_MAX;
1410
1411#if 0 /* disabled, the musl way to do things is just too racy */
1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1413 /* on linux we may be able to ask the kernel */
1414 {
1415 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1416
1417 if (fd >= 0)
733 { 1418 {
1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 close (fd);
1422
1423 /* here we should probably stat the open file and the disk file, to make sure they still match */
1424
1425 if (req->result > 0)
1426 goto done;
1427 }
1428 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1429 return;
1430 }
1431#endif
1432#endif
1433
1434 if (*rel != '/')
1435 {
1436 int len;
1437
734 errno = 0; 1438 errno = ENOENT;
735 readdir_r (dirp, &u->d, &entp); 1439 if (wd == EIO_INVALID_WD)
1440 return -1;
1441
1442 if (wd == EIO_CWD)
1443 {
1444 if (!getcwd (res, PATH_MAX))
1445 return -1;
736 1446
737 if (!entp) 1447 len = strlen (res);
1448 }
1449 else
1450 memcpy (res, wd->str, len = wd->len);
1451
1452 if (res [1]) /* only use if not / */
1453 res += len;
1454 }
1455
1456 while (*rel)
1457 {
1458 eio_ssize_t len, linklen;
1459 const char *beg = rel;
1460
1461 while (*rel && *rel != '/')
1462 ++rel;
1463
1464 len = rel - beg;
1465
1466 if (!len) /* skip slashes */
1467 {
738 break; 1468 ++rel;
1469 continue;
1470 }
739 1471
740 name = entp->d_name; 1472 if (beg [0] == '.')
1473 {
1474 if (len == 1)
1475 continue; /* . - nop */
741 1476
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1477 if (beg [1] == '.' && len == 2)
1478 {
1479 /* .. - back up one component, if possible */
1480
1481 while (res != tmpbuf->ptr)
1482 if (*--res == '/')
1483 break;
1484
1485 continue;
1486 }
1487 }
1488
1489 errno = ENAMETOOLONG;
1490 if (res + 1 + len + 1 >= tmp1)
1491 return -1;
1492
1493 /* copy one component */
1494 *res = '/';
1495 memcpy (res + 1, beg, len);
1496
1497 /* zero-terminate, for readlink */
1498 res [len + 1] = 0;
1499
1500 /* now check if it's a symlink */
1501 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1502
1503 if (linklen < 0)
743 { 1504 {
1505 if (errno != EINVAL)
1506 return -1;
1507
1508 /* it's a normal directory. hopefully */
1509 res += len + 1;
1510 }
1511 else
1512 {
1513 /* yay, it was a symlink - build new path in tmp2 */
744 int len = strlen (name) + 1; 1514 int rellen = strlen (rel);
745 1515
1516 errno = ENAMETOOLONG;
1517 if (linklen + 1 + rellen >= PATH_MAX)
1518 return -1;
1519
1520 errno = ELOOP;
1521 if (!--symlinks)
1522 return -1;
1523
1524 if (*tmp1 == '/')
1525 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1526
1527 /* we need to be careful, as rel might point into tmp2 already */
1528 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1529 tmp2 [linklen] = '/';
1530 memcpy (tmp2, tmp1, linklen);
1531
1532 rel = tmp2;
1533 }
1534 }
1535
1536 /* special case for the lone root path */
1537 if (res == tmpbuf->ptr)
1538 *res++ = '/';
1539
1540 return res - (char *)tmpbuf->ptr;
1541}
1542
1543static signed char
1544eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1545{
1546 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1547 : a->inode < b->inode ? -1
1548 : a->inode > b->inode ? 1
1549 : 0;
1550}
1551
1552#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1553
1554#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1555#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1556
1557static void
1558eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1559{
1560 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1561 unsigned char *bit = bits;
1562
1563 assert (CHAR_BIT == 8);
1564 assert (sizeof (eio_dirent) * 8 < 256);
1565 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1566 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1567
1568 if (size <= EIO_SORT_FAST)
1569 return;
1570
1571 /* first prepare an array of bits to test in our radix sort */
1572 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1573 /* inode_bits must contain all inodes ORed together */
1574 /* which is used to skip bits that are 0 everywhere, which is very common */
1575 {
1576 eio_ino_t endianness;
1577 int i, j;
1578
1579 /* we store the byte offset of byte n into byte n of "endianness" */
1580 for (i = 0; i < sizeof (eio_ino_t); ++i)
1581 ((unsigned char *)&endianness)[i] = i;
1582
1583 *bit++ = 0;
1584
1585 for (i = 0; i < sizeof (eio_ino_t); ++i)
1586 {
1587 /* shifting off the byte offsets out of "endianness" */
1588 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1589 endianness >>= 8;
1590
1591 for (j = 0; j < 8; ++j)
1592 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1593 *bit++ = offs + j;
1594 }
1595
1596 for (j = 0; j < 8; ++j)
1597 if (score_bits & (1 << j))
1598 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1599 }
1600
1601 /* now actually do the sorting (a variant of MSD radix sort) */
1602 {
1603 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1604 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1605 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1606 int stk_idx = 0;
1607
1608 base_stk [stk_idx] = dents;
1609 end_stk [stk_idx] = dents + size;
1610 bit_stk [stk_idx] = bit - 1;
1611
1612 do
1613 {
1614 base = base_stk [stk_idx];
1615 end = end_stk [stk_idx];
1616 bit = bit_stk [stk_idx];
1617
1618 for (;;)
1619 {
1620 unsigned char O = *bit >> 3;
1621 unsigned char M = 1 << (*bit & 7);
1622
1623 eio_dirent *a = base;
1624 eio_dirent *b = end;
1625
1626 if (b - a < EIO_SORT_CUTOFF)
746 res++; 1627 break;
747 1628
748 while (memofs + len > memlen) 1629 /* now bit-partition the array on the bit */
1630 /* this ugly asymmetric loop seems to perform much better than typical */
1631 /* partition algos found in the literature */
1632 do
1633 if (!(((unsigned char *)a)[O] & M))
1634 ++a;
1635 else if (!(((unsigned char *)--b)[O] & M))
749 { 1636 {
750 memlen *= 2; 1637 eio_dirent tmp = *a; *a = *b; *b = tmp;
751 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen);
753 X_UNLOCK (wrklock);
754
755 if (!names)
756 break; 1638 ++a;
757 } 1639 }
1640 while (b > a);
758 1641
759 memcpy (names + memofs, name, len); 1642 /* next bit, or stop, if no bits left in this path */
760 memofs += len; 1643 if (!*--bit)
1644 break;
1645
1646 base_stk [stk_idx] = a;
1647 end_stk [stk_idx] = end;
1648 bit_stk [stk_idx] = bit;
1649 ++stk_idx;
1650
1651 end = a;
761 } 1652 }
762 } 1653 }
763 1654 while (stk_idx--);
764 if (errno)
765 res = -1;
766 1655 }
1656}
1657
1658static void
1659eio_dent_insertion_sort (eio_dirent *dents, int size)
1660{
1661 /* first move the smallest element to the front, to act as a sentinel */
1662 {
1663 int i;
1664 eio_dirent *min = dents;
1665
1666 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1667 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1668 if (EIO_DENT_CMP (dents [i], <, *min))
1669 min = &dents [i];
1670
1671 /* swap elements 0 and j (minimum) */
1672 {
1673 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1674 }
1675 }
1676
1677 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1678 {
1679 eio_dirent *i, *j;
1680
1681 for (i = dents + 1; i < dents + size; ++i)
1682 {
1683 eio_dirent value = *i;
1684
1685 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1686 j [1] = j [0];
1687
1688 j [1] = value;
1689 }
1690 }
1691}
1692
1693static void
1694eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1695{
1696 if (size <= 1)
1697 return; /* our insertion sort relies on size > 0 */
1698
1699 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1700 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1701 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1702
1703 /* use an insertion sort at the end, or for small arrays, */
1704 /* as insertion sort is more efficient for small partitions */
1705 eio_dent_insertion_sort (dents, size);
1706}
1707
1708/* read a full directory */
1709static void
1710eio__scandir (eio_req *req, etp_worker *self)
1711{
1712 char *name, *names;
1713 int namesalloc = 4096 - sizeof (void *) * 4;
1714 int namesoffs = 0;
1715 int flags = req->int1;
1716 eio_dirent *dents = 0;
1717 int dentalloc = 128;
1718 int dentoffs = 0;
1719 eio_ino_t inode_bits = 0;
1720#ifdef _WIN32
1721 HANDLE dirp;
1722 WIN32_FIND_DATA entp;
1723#else
1724 DIR *dirp;
1725 EIO_STRUCT_DIRENT *entp;
1726#endif
1727
767 req->result = res; 1728 req->result = -1;
1729
1730 if (!(flags & EIO_READDIR_DENTS))
1731 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1732
1733#ifdef _WIN32
1734 {
1735 int len = strlen ((const char *)req->ptr1);
1736 char *path = malloc (MAX_PATH);
1737 const char *fmt;
1738 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1739
1740 if (!len)
1741 fmt = "./*";
1742 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1743 fmt = "%s*";
1744 else
1745 fmt = "%s/*";
1746
1747 _snprintf (path, MAX_PATH, fmt, reqpath);
1748 dirp = FindFirstFile (path, &entp);
1749 free (path);
1750
1751 if (dirp == INVALID_HANDLE_VALUE)
1752 {
1753 /* should steal _dosmaperr */
1754 switch (GetLastError ())
1755 {
1756 case ERROR_FILE_NOT_FOUND:
1757 req->result = 0;
1758 break;
1759
1760 case ERROR_INVALID_NAME:
1761 case ERROR_PATH_NOT_FOUND:
1762 case ERROR_NO_MORE_FILES:
1763 errno = ENOENT;
1764 break;
1765
1766 case ERROR_NOT_ENOUGH_MEMORY:
1767 errno = ENOMEM;
1768 break;
1769
1770 default:
1771 errno = EINVAL;
1772 break;
1773 }
1774
1775 return;
1776 }
1777 }
1778#else
1779 #if HAVE_AT
1780 if (req->wd)
1781 {
1782 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1783
1784 if (fd < 0)
1785 return;
1786
1787 dirp = fdopendir (fd);
1788
1789 if (!dirp)
1790 close (fd);
1791 }
1792 else
1793 dirp = opendir (req->ptr1);
1794 #else
1795 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1796 #endif
1797
1798 if (!dirp)
1799 return;
1800#endif
1801
1802 if (req->flags & EIO_FLAG_PTR1_FREE)
1803 free (req->ptr1);
1804
1805 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1806 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1807 req->ptr2 = names = malloc (namesalloc);
1808
1809 if (!names || (flags && !dents))
1810 return;
1811
1812 for (;;)
1813 {
1814 int done;
1815
1816#ifdef _WIN32
1817 done = !dirp;
1818#else
1819 errno = 0;
1820 entp = readdir (dirp);
1821 done = !entp;
1822#endif
1823
1824 if (done)
1825 {
1826#ifndef _WIN32
1827 int old_errno = errno;
1828 closedir (dirp);
1829 errno = old_errno;
1830
1831 if (errno)
1832 break;
1833#endif
1834
1835 /* sort etc. */
1836 req->int1 = flags;
1837 req->result = dentoffs;
1838
1839 if (flags & EIO_READDIR_STAT_ORDER)
1840 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1841 else if (flags & EIO_READDIR_DIRS_FIRST)
1842 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1843 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1844 else
1845 {
1846 /* in this case, all is known, and we just put dirs first and sort them */
1847 eio_dirent *oth = dents + dentoffs;
1848 eio_dirent *dir = dents;
1849
1850 /* now partition dirs to the front, and non-dirs to the back */
1851 /* by walking from both sides and swapping if necessary */
1852 while (oth > dir)
1853 {
1854 if (dir->type == EIO_DT_DIR)
1855 ++dir;
1856 else if ((--oth)->type == EIO_DT_DIR)
1857 {
1858 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1859
1860 ++dir;
1861 }
1862 }
1863
1864 /* now sort the dirs only (dirs all have the same score) */
1865 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1866 }
1867
1868 break;
1869 }
1870
1871 /* now add the entry to our list(s) */
1872 name = D_NAME (entp);
1873
1874 /* skip . and .. entries */
1875 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1876 {
1877 int len = D_NAMLEN (entp) + 1;
1878
1879 while (ecb_expect_false (namesoffs + len > namesalloc))
1880 {
1881 namesalloc *= 2;
1882 req->ptr2 = names = realloc (names, namesalloc);
1883
1884 if (!names)
1885 break;
1886 }
1887
1888 memcpy (names + namesoffs, name, len);
1889
1890 if (dents)
1891 {
1892 struct eio_dirent *ent;
1893
1894 if (ecb_expect_false (dentoffs == dentalloc))
1895 {
1896 dentalloc *= 2;
1897 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1898
1899 if (!dents)
1900 break;
1901 }
1902
1903 ent = dents + dentoffs;
1904
1905 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1906 ent->namelen = len - 1;
1907 ent->inode = D_INO (entp);
1908
1909 inode_bits |= ent->inode;
1910
1911 switch (D_TYPE (entp))
1912 {
1913 default:
1914 ent->type = EIO_DT_UNKNOWN;
1915 flags |= EIO_READDIR_FOUND_UNKNOWN;
1916 break;
1917
1918 #ifdef DT_FIFO
1919 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1920 #endif
1921 #ifdef DT_CHR
1922 case DT_CHR: ent->type = EIO_DT_CHR; break;
1923 #endif
1924 #ifdef DT_MPC
1925 case DT_MPC: ent->type = EIO_DT_MPC; break;
1926 #endif
1927 #ifdef DT_DIR
1928 case DT_DIR: ent->type = EIO_DT_DIR; break;
1929 #endif
1930 #ifdef DT_NAM
1931 case DT_NAM: ent->type = EIO_DT_NAM; break;
1932 #endif
1933 #ifdef DT_BLK
1934 case DT_BLK: ent->type = EIO_DT_BLK; break;
1935 #endif
1936 #ifdef DT_MPB
1937 case DT_MPB: ent->type = EIO_DT_MPB; break;
1938 #endif
1939 #ifdef DT_REG
1940 case DT_REG: ent->type = EIO_DT_REG; break;
1941 #endif
1942 #ifdef DT_NWK
1943 case DT_NWK: ent->type = EIO_DT_NWK; break;
1944 #endif
1945 #ifdef DT_CMP
1946 case DT_CMP: ent->type = EIO_DT_CMP; break;
1947 #endif
1948 #ifdef DT_LNK
1949 case DT_LNK: ent->type = EIO_DT_LNK; break;
1950 #endif
1951 #ifdef DT_SOCK
1952 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1953 #endif
1954 #ifdef DT_DOOR
1955 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1956 #endif
1957 #ifdef DT_WHT
1958 case DT_WHT: ent->type = EIO_DT_WHT; break;
1959 #endif
1960 }
1961
1962 ent->score = 7;
1963
1964 if (flags & EIO_READDIR_DIRS_FIRST)
1965 {
1966 if (ent->type == EIO_DT_UNKNOWN)
1967 {
1968 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1969 ent->score = 1;
1970 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1971 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1972 }
1973 else if (ent->type == EIO_DT_DIR)
1974 ent->score = 0;
1975 }
1976 }
1977
1978 namesoffs += len;
1979 ++dentoffs;
1980 }
1981
1982 if (EIO_CANCELLED (req))
1983 {
1984 errno = ECANCELED;
1985 break;
1986 }
1987
1988#ifdef _WIN32
1989 if (!FindNextFile (dirp, &entp))
1990 {
1991 FindClose (dirp);
1992 dirp = 0;
1993 }
1994#endif
1995 }
768} 1996}
1997
1998/*****************************************************************************/
1999/* working directory stuff */
2000/* various deficiencies in the posix 2008 api force us to */
2001/* keep the absolute path in string form at all times */
2002/* fuck yeah. */
2003
2004#if !HAVE_AT
2005
2006/* a bit like realpath, but usually faster because it doesn'T have to return */
2007/* an absolute or canonical path */
2008static const char *
2009wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2010{
2011 if (!wd || *path == '/')
2012 return path;
2013
2014 if (path [0] == '.' && !path [1])
2015 return wd->str;
2016
2017 {
2018 int l1 = wd->len;
2019 int l2 = strlen (path);
2020
2021 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2022
2023 memcpy (res, wd->str, l1);
2024 res [l1] = '/';
2025 memcpy (res + l1 + 1, path, l2 + 1);
2026
2027 return res;
2028 }
2029}
2030
2031#endif
2032
2033static eio_wd
2034eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2035{
2036 int fd;
2037 eio_wd res;
2038 int len = eio__realpath (tmpbuf, wd, path);
2039
2040 if (len < 0)
2041 return EIO_INVALID_WD;
2042
2043#if HAVE_AT
2044 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2045
2046 if (fd < 0)
2047 return EIO_INVALID_WD;
2048#endif
2049
2050 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2051
2052#if HAVE_AT
2053 res->fd = fd;
2054#endif
2055
2056 res->len = len;
2057 memcpy (res->str, tmpbuf->ptr, len);
2058 res->str [len] = 0;
2059
2060 return res;
2061}
2062
2063eio_wd
2064eio_wd_open_sync (eio_wd wd, const char *path)
2065{
2066 struct tmpbuf tmpbuf = { 0 };
2067 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2068 free (tmpbuf.ptr);
2069
2070 return wd;
2071}
2072
2073void
2074eio_wd_close_sync (eio_wd wd)
2075{
2076 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2077 {
2078 #if HAVE_AT
2079 close (wd->fd);
2080 #endif
2081 free (wd);
2082 }
2083}
2084
2085#if HAVE_AT
2086
2087/* they forgot these */
2088
2089static int
2090eio__truncateat (int dirfd, const char *path, off_t length)
2091{
2092 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2093 int res;
2094
2095 if (fd < 0)
2096 return fd;
2097
2098 res = ftruncate (fd, length);
2099 close (fd);
2100 return res;
2101}
2102
2103static int
2104eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2105{
2106 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2107 int res;
2108
2109 if (fd < 0)
2110 return fd;
2111
2112 res = fstatvfs (fd, buf);
2113 close (fd);
2114 return res;
2115
2116}
2117
2118#endif
769 2119
770/*****************************************************************************/ 2120/*****************************************************************************/
771 2121
772#define ALLOC(len) \ 2122#define ALLOC(len) \
773 if (!req->ptr2) \ 2123 if (!req->ptr2) \
… …
782 req->result = -1; \ 2132 req->result = -1; \
783 break; \ 2133 break; \
784 } \ 2134 } \
785 } 2135 }
786 2136
2137static void ecb_noinline ecb_cold
2138etp_proc_init (void)
2139{
2140#if HAVE_PRCTL_SET_NAME
2141 /* provide a more sensible "thread name" */
2142 char name[16 + 1];
2143 const int namelen = sizeof (name) - 1;
2144 int len;
2145
2146 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2147 name [namelen] = 0;
2148 len = strlen (name);
2149 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2150 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2151#endif
2152}
2153
787X_THREAD_PROC (eio_proc) 2154X_THREAD_PROC (etp_proc)
788{ 2155{
789 eio_req *req; 2156 ETP_REQ *req;
790 struct timespec ts; 2157 struct timespec ts;
791 worker *self = (worker *)thr_arg; 2158 etp_worker *self = (etp_worker *)thr_arg;
792 2159
2160 etp_proc_init ();
2161
793 /* try to distribute timeouts somewhat randomly */ 2162 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2163 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 2164
796 for (;;) 2165 for (;;)
797 { 2166 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2167 ts.tv_sec = 0;
799 2168
800 X_LOCK (reqlock); 2169 X_LOCK (reqlock);
801 2170
802 for (;;) 2171 for (;;)
803 { 2172 {
804 self->req = req = reqq_shift (&req_queue); 2173 req = reqq_shift (&req_queue);
805 2174
806 if (req) 2175 if (req)
807 break; 2176 break;
808 2177
2178 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2179 {
2180 X_UNLOCK (reqlock);
2181 X_LOCK (wrklock);
2182 --started;
2183 X_UNLOCK (wrklock);
2184 goto quit;
2185 }
2186
809 ++idle; 2187 ++idle;
810 2188
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2189 if (idle <= max_idle)
2190 /* we are allowed to idle, so do so without any timeout */
2191 X_COND_WAIT (reqwait, reqlock);
2192 else
812 { 2193 {
813 if (idle > max_idle) 2194 /* initialise timeout once */
814 { 2195 if (!ts.tv_sec)
815 --idle; 2196 ts.tv_sec = time (0) + idle_timeout;
816 X_UNLOCK (reqlock);
817 X_LOCK (wrklock);
818 --started;
819 X_UNLOCK (wrklock);
820 goto quit;
821 }
822 2197
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 2198 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2199 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 2200 }
827 2201
828 --idle; 2202 --idle;
829 } 2203 }
830 2204
831 --nready; 2205 --nready;
832 2206
833 X_UNLOCK (reqlock); 2207 X_UNLOCK (reqlock);
834 2208
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 2209 if (req->type < 0)
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit; 2210 goto quit;
930 2211
931 default: 2212 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 2213
938 X_LOCK (reslock); 2214 X_LOCK (reslock);
939 2215
940 ++npending; 2216 ++npending;
941 2217
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 2218 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 2219 want_poll_cb ();
944 2220
945 self->req = 0;
946 worker_clear (self); 2221 etp_worker_clear (self);
947 2222
948 X_UNLOCK (reslock); 2223 X_UNLOCK (reslock);
949 } 2224 }
950 2225
951quit: 2226quit:
2227 free (req);
2228
952 X_LOCK (wrklock); 2229 X_LOCK (wrklock);
953 worker_free (self); 2230 etp_worker_free (self);
954 X_UNLOCK (wrklock); 2231 X_UNLOCK (wrklock);
955 2232
956 return 0; 2233 return 0;
957} 2234}
958 2235
959/*****************************************************************************/ 2236/*****************************************************************************/
960 2237
961static void eio_atfork_prepare (void) 2238int ecb_cold
2239eio_init (void (*want_poll)(void), void (*done_poll)(void))
962{ 2240{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE 2241#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{
1019 want_poll_cb = want_poll;
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock); 2242 X_MUTEX_CREATE (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif 2243#endif
1032 2244
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2245 return etp_init (want_poll, done_poll);
1034} 2246}
1035 2247
2248ecb_inline void
1036static void eio_api_destroy (eio_req *req) 2249eio_api_destroy (eio_req *req)
1037{ 2250{
1038 free (req); 2251 free (req);
1039} 2252}
1040 2253
1041#define REQ(rtype) \ 2254#define REQ(rtype) \
… …
1043 \ 2256 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 2257 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 2258 if (!req) \
1046 return 0; \ 2259 return 0; \
1047 \ 2260 \
1048 req->type = EIO_ ## rtype; \ 2261 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 2262 req->pri = pri; \
1050 req->finish = cb; \ 2263 req->finish = cb; \
2264 req->data = data; \
1051 req->destroy = eio_api_destroy; 2265 req->destroy = eio_api_destroy;
1052 2266
1053#define SEND eio_submit (req); return req 2267#define SEND eio_submit (req); return req
1054 2268
1055#define PATH (void)0 2269#define PATH \
2270 req->flags |= EIO_FLAG_PTR1_FREE; \
2271 req->ptr1 = strdup (path); \
2272 if (!req->ptr1) \
2273 { \
2274 eio_api_destroy (req); \
2275 return 0; \
2276 }
1056 2277
2278static void
2279eio_execute (etp_worker *self, eio_req *req)
2280{
2281#if HAVE_AT
2282 int dirfd;
2283#else
2284 const char *path;
2285#endif
2286
2287 if (ecb_expect_false (EIO_CANCELLED (req)))
2288 {
2289 req->result = -1;
2290 req->errorno = ECANCELED;
2291 return;
2292 }
2293
2294 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2295 {
2296 req->result = -1;
2297 req->errorno = ENOENT;
2298 return;
2299 }
2300
2301 if (req->type >= EIO_OPEN)
2302 {
2303 #if HAVE_AT
2304 dirfd = WD2FD (req->wd);
2305 #else
2306 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2307 #endif
2308 }
2309
2310 switch (req->type)
2311 {
2312 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2313 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2314 break;
2315 case EIO_WD_CLOSE: req->result = 0;
2316 eio_wd_close_sync (req->wd); break;
2317
2318 case EIO_READ: ALLOC (req->size);
2319 req->result = req->offs >= 0
2320 ? pread (req->int1, req->ptr2, req->size, req->offs)
2321 : read (req->int1, req->ptr2, req->size); break;
2322 case EIO_WRITE: req->result = req->offs >= 0
2323 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2324 : write (req->int1, req->ptr2, req->size); break;
2325
2326 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2327 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2328
2329#if HAVE_AT
2330
2331 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2333 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2334 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2335 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2336 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2337 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2338 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2339
2340 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2341 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2342 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2343 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2344 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2345 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2346 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2347 case EIO_READLINK: ALLOC (PATH_MAX);
2348 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2349 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2350 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2351 case EIO_UTIME:
2352 case EIO_FUTIME:
2353 {
2354 struct timespec ts[2];
2355 struct timespec *times;
2356
2357 if (req->nv1 != -1. || req->nv2 != -1.)
2358 {
2359 ts[0].tv_sec = req->nv1;
2360 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2361 ts[1].tv_sec = req->nv2;
2362 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2363
2364 times = ts;
2365 }
2366 else
2367 times = 0;
2368
2369 req->result = req->type == EIO_FUTIME
2370 ? futimens (req->int1, times)
2371 : utimensat (dirfd, req->ptr1, times, 0);
2372 }
2373 break;
2374
2375#else
2376
2377 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2378 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2379 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2380 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2381 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2382 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2383 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2384 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2385
2386 case EIO_UNLINK: req->result = unlink (path ); break;
2387 case EIO_RMDIR: req->result = rmdir (path ); break;
2388 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2389 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2390 case EIO_LINK: req->result = link (path , req->ptr2); break;
2391 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2392 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2393 case EIO_READLINK: ALLOC (PATH_MAX);
2394 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2395 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2396 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2397
2398 case EIO_UTIME:
2399 case EIO_FUTIME:
2400 {
2401 struct timeval tv[2];
2402 struct timeval *times;
2403
2404 if (req->nv1 != -1. || req->nv2 != -1.)
2405 {
2406 tv[0].tv_sec = req->nv1;
2407 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2408 tv[1].tv_sec = req->nv2;
2409 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2410
2411 times = tv;
2412 }
2413 else
2414 times = 0;
2415
2416 req->result = req->type == EIO_FUTIME
2417 ? futimes (req->int1, times)
2418 : utimes (req->ptr1, times);
2419 }
2420 break;
2421
2422#endif
2423
2424 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2425 {
2426 ALLOC (req->result);
2427 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2428 }
2429 break;
2430
2431 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2432 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2433
2434 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2435 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2436
2437 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2438 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2439 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2440
2441 case EIO_CLOSE: req->result = close (req->int1); break;
2442 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2443 case EIO_SYNC: req->result = 0; sync (); break;
2444 case EIO_FSYNC: req->result = fsync (req->int1); break;
2445 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2446 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2447 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2448 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2449 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2450 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2451 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2452 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2453
2454 case EIO_READDIR: eio__scandir (req, self); break;
2455
2456 case EIO_BUSY:
2457#ifdef _WIN32
2458 Sleep (req->nv1 * 1e3);
2459#else
2460 {
2461 struct timeval tv;
2462
2463 tv.tv_sec = req->nv1;
2464 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2465
2466 req->result = select (0, 0, 0, 0, &tv);
2467 }
2468#endif
2469 break;
2470
2471 case EIO_GROUP:
2472 abort (); /* handled in eio_request */
2473
2474 case EIO_NOP:
2475 req->result = 0;
2476 break;
2477
2478 case EIO_CUSTOM:
2479 req->feed (req);
2480 break;
2481
2482 default:
2483 errno = ENOSYS;
2484 req->result = -1;
2485 break;
2486 }
2487
2488 req->errorno = errno;
2489}
2490
2491#ifndef EIO_NO_WRAPPERS
2492
2493eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2494{
2495 REQ (EIO_WD_OPEN); PATH; SEND;
2496}
2497
2498eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2499{
2500 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2501}
2502
2503eio_req *eio_nop (int pri, eio_cb cb, void *data)
2504{
2505 REQ (EIO_NOP); SEND;
2506}
2507
2508eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2509{
2510 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2511}
2512
1057eio_req *eio_fsync (int fd, eio_cb cb) 2513eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2514{
2515 REQ (EIO_SYNC); SEND;
2516}
2517
2518eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2519{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2520 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2521}
1061 2522
2523eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2524{
2525 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2526}
2527
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2528eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2529{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2530 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2531}
1066 2532
2533eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2534{
2535 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2536}
2537
2538eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2539{
2540 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2541}
2542
2543eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2544{
2545 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2546}
2547
2548eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2549{
2550 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2551}
2552
2553eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2554{
2555 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2556}
2557
2558eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2559{
2560 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2561}
2562
2563eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2564{
2565 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2566}
2567
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2568eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2569{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2570 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2571}
1071 2572
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2573eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2574{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2575 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2576}
1076 2577
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2578eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2579{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2580 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2581}
1081 2582
1082eio_req *eio_fstat (int fd, eio_cb cb) 2583eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2584{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2585 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2586}
1086 2587
2588eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2589{
2590 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2591}
2592
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2593eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2594{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2595 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2596}
1091 2597
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2598eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2599{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2600 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2601}
1096 2602
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2603eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2604{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2605 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2606}
1101 2607
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2608eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2609{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2610 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2611}
1106 2612
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2613eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2614{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2615 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2616}
1111 2617
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2618eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2619{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2620 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2621}
1116 2622
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2623eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2624{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2625 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2626}
1121 2627
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2628eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2629{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2630 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2631}
1141 2632
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2633eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2634{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2635 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2636}
1146 2637
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2638eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2639{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2640 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2641}
1151 2642
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2643eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2644{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2645 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2646}
1156 2647
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2648eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2649{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2650 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2651}
1161 2652
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2653static eio_req *
2654eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2655{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2656 REQ (type); PATH; SEND;
1170} 2657}
1171 2658
2659eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2660{
2661 return eio__1path (EIO_READLINK, path, pri, cb, data);
2662}
2663
2664eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2665{
2666 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2667}
2668
2669eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2670{
2671 return eio__1path (EIO_STAT, path, pri, cb, data);
2672}
2673
2674eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2675{
2676 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2677}
2678
2679eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2680{
2681 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2682}
2683
2684eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2685{
2686 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2687}
2688
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2689eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2690{
1174 REQ (READDIR); PATH; SEND; 2691 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2692}
1176 2693
2694eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2695{
2696 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2697}
2698
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2699eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2700{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2701 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2702}
1181 2703
1182eio_req *eio_busy (double delay, eio_cb cb) 2704static eio_req *
2705eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2706{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2707 REQ (type); PATH;
1185}
1186 2708
2709 req->flags |= EIO_FLAG_PTR2_FREE;
2710 req->ptr2 = strdup (new_path);
2711 if (!req->ptr2)
2712 {
2713 eio_api_destroy (req);
2714 return 0;
2715 }
2716
2717 SEND;
2718}
2719
2720eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2721{
2722 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2723}
2724
2725eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2726{
2727 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2728}
2729
2730eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2731{
2732 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2733}
2734
2735eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2736{
2737 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2738}
2739
2740#endif
2741
1187eio_req *eio_nop (eio_cb cb) 2742eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2743{
2744 const int pri = EIO_PRI_MAX;
2745
1189 REQ (NOP); SEND; 2746 REQ (EIO_GROUP); SEND;
1190} 2747}
1191 2748
1192#undef REQ 2749#undef REQ
1193#undef PATH 2750#undef PATH
1194#undef SEND 2751#undef SEND
1195 2752
1196#if 0 2753/*****************************************************************************/
2754/* grp functions */
1197 2755
1198void 2756void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2757eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2758{
1573 grp->int2 = limit; 2759 grp->int2 = limit;
1574 grp->feed = feed; 2760 grp->feed = feed;
1575 2761
1576 grp_try_feed (grp); 2762 grp_try_feed (grp);
1577} 2763}
1578 2764
2765void
1579void eio_grp_limit (eio_req *grp, int limit) 2766eio_grp_limit (eio_req *grp, int limit)
1580{ 2767{
1581 grp->int2 = limit; 2768 grp->int2 = limit;
1582 2769
1583 grp_try_feed (grp); 2770 grp_try_feed (grp);
1584} 2771}
1585 2772
2773void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2774eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2775{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2776 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2777
2778 grp->flags |= EIO_FLAG_GROUPADD;
1589 2779
1590 ++grp->size; 2780 ++grp->size;
1591 req->grp = grp; 2781 req->grp = grp;
1592 2782
1593 req->grp_prev = 0; 2783 req->grp_prev = 0;
… …
1597 grp->grp_first->grp_prev = req; 2787 grp->grp_first->grp_prev = req;
1598 2788
1599 grp->grp_first = req; 2789 grp->grp_first = req;
1600} 2790}
1601 2791
2792/*****************************************************************************/
2793/* misc garbage */
1602 2794
2795eio_ssize_t
2796eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2797{
2798 return eio__sendfile (ofd, ifd, offset, count);
2799}
2800

Diff Legend

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