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

Comparing libeio/eio.c (file contents):
Revision 1.4 by root, Sun May 11 00:10:15 2008 UTC vs.
Revision 1.98 by root, Sun Jul 24 05:53:34 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
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);
14 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
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 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 162
31#else 163#else
32 164
33# include "config.h"
34# include <sys/time.h> 165 #include <sys/time.h>
35# include <sys/select.h> 166 #include <sys/select.h>
167 #include <sys/statvfs.h>
36# include <unistd.h> 168 #include <unistd.h>
169 #include <signal.h>
170 #include <dirent.h>
171
172 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
173 #include <sys/mman.h>
174 #endif
175
176 #define D_NAME(entp) entp->d_name
177
178 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
180 #define _DIRENT_HAVE_D_TYPE /* sigh */
181 #define D_INO(de) (de)->d_fileno
182 #define D_NAMLEN(de) (de)->d_namlen
183 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184 #define D_INO(de) (de)->d_ino
185 #endif
186
187 #ifdef _D_EXACT_NAMLEN
188 #undef D_NAMLEN
189 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
190 #endif
191
192 #ifdef _DIRENT_HAVE_D_TYPE
193 #define D_TYPE(de) (de)->d_type
194 #endif
195
196 #ifndef EIO_STRUCT_DIRENT
197 #define EIO_STRUCT_DIRENT struct dirent
198 #endif
199
200#endif
201
202#if HAVE_UTIMES
37# include <utime.h> 203# 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 204#endif
44 205
45#endif 206#if HAVE_SYS_SYSCALL_H
46 207# include <sys/syscall.h>
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif 208#endif
209
210#if HAVE_SYS_PRCTL_H
211# include <sys/prctl.h>
212#endif
50 213
51#if HAVE_SENDFILE 214#if HAVE_SENDFILE
52# if __linux 215# if __linux
53# include <sys/sendfile.h> 216# include <sys/sendfile.h>
54# elif __freebsd 217# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 218# include <sys/socket.h>
56# include <sys/uio.h> 219# include <sys/uio.h>
57# elif __hpux 220# elif __hpux
58# include <sys/socket.h> 221# include <sys/socket.h>
59# elif __solaris /* not yet */ 222# elif __solaris
60# include <sys/sendfile.h> 223# include <sys/sendfile.h>
61# else 224# else
62# error sendfile support requested but not available 225# error sendfile support requested but not available
63# endif 226# endif
64#endif 227#endif
65 228
66/* number of seconds after which an idle threads exit */ 229#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 230# define D_TYPE(de) 0
231#endif
232#ifndef D_INO
233# define D_INO(de) 0
234#endif
235#ifndef D_NAMLEN
236# define D_NAMLEN(entp) strlen (D_NAME (entp))
237#endif
68 238
69/* used for struct dirent, AIX doesn't provide it */ 239/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 240#ifndef NAME_MAX
71# define NAME_MAX 4096 241# define NAME_MAX 4096
72#endif 242#endif
73 243
244/* used for readlink etc. */
245#ifndef PATH_MAX
246# define PATH_MAX 4096
247#endif
248
74/* buffer size for various temporary buffers */ 249/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 250#define EIO_BUFSIZE 65536
76 251
77#define dBUF \ 252#define dBUF \
78 char *eio_buf; \
79 X_LOCK (wrklock); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 253 char *eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 254 errno = ENOMEM; \
82 if (!eio_buf) \ 255 if (!eio_buf) \
83 return -1; 256 return -1
257
258#define FUBd \
259 free (eio_buf)
84 260
85#define EIO_TICKS ((1000000 + 1023) >> 10) 261#define EIO_TICKS ((1000000 + 1023) >> 10)
262
263#define ETP_PRI_MIN EIO_PRI_MIN
264#define ETP_PRI_MAX EIO_PRI_MAX
265
266struct etp_worker;
267
268#define ETP_REQ eio_req
269#define ETP_DESTROY(req) eio_destroy (req)
270static int eio_finish (eio_req *req);
271#define ETP_FINISH(req) eio_finish (req)
272static void eio_execute (struct etp_worker *self, eio_req *req);
273#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
274
275/*****************************************************************************/
276
277#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
278
279/* calculate time difference in ~1/EIO_TICKS of a second */
280ecb_inline int
281tvdiff (struct timeval *tv1, struct timeval *tv2)
282{
283 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
284 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
285}
286
287static unsigned int started, idle, wanted = 4;
86 288
87static void (*want_poll_cb) (void); 289static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 290static void (*done_poll_cb) (void);
89 291
90static unsigned int max_poll_time = 0; 292static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 293static unsigned int max_poll_reqs; /* reslock */
92 294
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 295static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 296static volatile unsigned int nready; /* reqlock */
95{ 297static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 298static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 299static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 300
100static unsigned int started, idle, wanted; 301static xmutex_t wrklock;
302static xmutex_t reslock;
303static xmutex_t reqlock;
304static xcond_t reqwait;
305
306#if !HAVE_PREADWRITE
307/*
308 * make our pread/pwrite emulation safe against themselves, but not against
309 * normal read/write by using a mutex. slows down execution a lot,
310 * but that's your problem, not mine.
311 */
312static xmutex_t preadwritelock;
313#endif
314
315typedef struct etp_worker
316{
317 /* locked by wrklock */
318 struct etp_worker *prev, *next;
319
320 xthread_t tid;
321
322 /* locked by reslock, reqlock or wrklock */
323 ETP_REQ *req; /* currently processed request */
324
325#ifdef ETP_WORKER_COMMON
326 ETP_WORKER_COMMON
327#endif
328} etp_worker;
329
330static etp_worker wrk_first; /* NOT etp */
331
332#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
333#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 334
102/* worker threads management */ 335/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 336
105typedef struct worker { 337static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 338etp_worker_clear (etp_worker *wrk)
120{ 339{
121 if (wrk->dirp)
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126
127 if (wrk->dbuf)
128 {
129 free (wrk->dbuf);
130 wrk->dbuf = 0;
131 }
132} 340}
133 341
134static void worker_free (worker *wrk) 342static void ecb_cold
343etp_worker_free (etp_worker *wrk)
135{ 344{
136 wrk->next->prev = wrk->prev; 345 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 346 wrk->prev->next = wrk->next;
138 347
139 free (wrk); 348 free (wrk);
140} 349}
141 350
142static volatile unsigned int nreqs, nready, npending; 351static unsigned int
143static volatile unsigned int max_idle = 4; 352etp_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{ 353{
354 int retval;
355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
356 retval = nreqs;
357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 358 return retval;
152} 359}
153 360
154unsigned int eio_nready (void) 361static unsigned int
362etp_nready (void)
155{ 363{
156 unsigned int retval; 364 unsigned int retval;
157 365
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 366 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 367 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 368 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 369
162 return retval; 370 return retval;
163} 371}
164 372
165unsigned int eio_npending (void) 373static unsigned int
374etp_npending (void)
166{ 375{
167 unsigned int retval; 376 unsigned int retval;
168 377
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 378 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 379 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 380 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 381
173 return retval; 382 return retval;
174} 383}
175 384
176unsigned int eio_nthreads (void) 385static unsigned int
386etp_nthreads (void)
177{ 387{
178 unsigned int retval; 388 unsigned int retval;
179 389
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 390 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 391 retval = started;
188 * a somewhat faster data structure might be nice, but 398 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 399 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 400 * per shift, the most expensive operation.
191 */ 401 */
192typedef struct { 402typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 403 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 404 int size;
195} reqq; 405} etp_reqq;
196 406
197static reqq req_queue; 407static etp_reqq req_queue;
198static reqq res_queue; 408static etp_reqq res_queue;
199 409
200static int reqq_push (reqq *q, eio_req *req) 410static void ecb_noinline ecb_cold
411reqq_init (etp_reqq *q)
412{
413 int pri;
414
415 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
416 q->qs[pri] = q->qe[pri] = 0;
417
418 q->size = 0;
419}
420
421static int ecb_noinline
422reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 423{
202 int pri = req->pri; 424 int pri = req->pri;
203 req->next = 0; 425 req->next = 0;
204 426
205 if (q->qe[pri]) 427 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 433 q->qe[pri] = q->qs[pri] = req;
212 434
213 return q->size++; 435 return q->size++;
214} 436}
215 437
216static eio_req *reqq_shift (reqq *q) 438static ETP_REQ * ecb_noinline
439reqq_shift (etp_reqq *q)
217{ 440{
218 int pri; 441 int pri;
219 442
220 if (!q->size) 443 if (!q->size)
221 return 0; 444 return 0;
222 445
223 --q->size; 446 --q->size;
224 447
225 for (pri = EIO_NUM_PRI; pri--; ) 448 for (pri = ETP_NUM_PRI; pri--; )
226 { 449 {
227 eio_req *req = q->qs[pri]; 450 eio_req *req = q->qs[pri];
228 451
229 if (req) 452 if (req)
230 { 453 {
236 } 459 }
237 460
238 abort (); 461 abort ();
239} 462}
240 463
241static void grp_try_feed (eio_req *grp) 464static int ecb_cold
465etp_init (void (*want_poll)(void), void (*done_poll)(void))
242{ 466{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 467 X_MUTEX_CREATE (wrklock);
244 { 468 X_MUTEX_CREATE (reslock);
245 int old_len = grp->size; 469 X_MUTEX_CREATE (reqlock);
470 X_COND_CREATE (reqwait);
246 471
247 EIO_FEED (grp); 472 reqq_init (&req_queue);
473 reqq_init (&res_queue);
248 474
249 /* stop if no progress has been made */ 475 wrk_first.next =
250 if (old_len == grp->size) 476 wrk_first.prev = &wrk_first;
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256}
257 477
258static int eio_finish (eio_req *req); 478 started = 0;
479 idle = 0;
480 nreqs = 0;
481 nready = 0;
482 npending = 0;
259 483
260static int grp_dec (eio_req *grp) 484 want_poll_cb = want_poll;
261{ 485 done_poll_cb = done_poll;
262 --grp->size;
263 486
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; 487 return 0;
272} 488}
273 489
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR2_FREE)
277 free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 req->flags |= EIO_FLAG_CANCELLED;
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 490X_THREAD_PROC (etp_proc);
323 491
324static void start_thread (void) 492static void ecb_cold
493etp_start_thread (void)
325{ 494{
326 worker *wrk = calloc (1, sizeof (worker)); 495 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 496
328 if (!wrk) 497 /*TODO*/
329 croak ("unable to allocate worker thread data"); 498 assert (("unable to allocate worker thread data", wrk));
330 499
331 X_LOCK (wrklock); 500 X_LOCK (wrklock);
332 501
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 502 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 503 {
335 wrk->prev = &wrk_first; 504 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 505 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 506 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 507 wrk_first.next = wrk;
342 free (wrk); 511 free (wrk);
343 512
344 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
345} 514}
346 515
516static void
347static void maybe_start_thread (void) 517etp_maybe_start_thread (void)
348{ 518{
349 if (eio_nthreads () >= wanted) 519 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 520 return;
351 521
352 /* todo: maybe use idle here, but might be less exact */ 522 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 523 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 524 return;
355 525
356 start_thread (); 526 etp_start_thread ();
357} 527}
358 528
359void eio_submit (eio_req *req) 529static void ecb_cold
360{ 530etp_end_thread (void)
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void)
373{ 531{
374 eio_req *req = calloc (1, sizeof (eio_req)); 532 eio_req *req = calloc (1, sizeof (eio_req));
375 533
376 req->type = EIO_QUIT; 534 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 535 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 536
379 X_LOCK (reqlock); 537 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 538 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 539 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 540 X_UNLOCK (reqlock);
384 X_LOCK (wrklock); 542 X_LOCK (wrklock);
385 --started; 543 --started;
386 X_UNLOCK (wrklock); 544 X_UNLOCK (wrklock);
387} 545}
388 546
389void eio_set_max_poll_time (double nseconds) 547static int
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 548etp_poll (void)
426{ 549{
427 int maxreqs = max_poll_reqs; 550 unsigned int maxreqs;
551 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 552 struct timeval tv_start, tv_now;
429 eio_req *req;
430 553
554 X_LOCK (reslock);
555 maxreqs = max_poll_reqs;
556 maxtime = max_poll_time;
557 X_UNLOCK (reslock);
558
431 if (max_poll_time) 559 if (maxtime)
432 gettimeofday (&tv_start, 0); 560 gettimeofday (&tv_start, 0);
433 561
434 for (;;) 562 for (;;)
435 { 563 {
564 ETP_REQ *req;
565
436 maybe_start_thread (); 566 etp_maybe_start_thread ();
437 567
438 X_LOCK (reslock); 568 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 569 req = reqq_shift (&res_queue);
440 570
441 if (req) 571 if (req)
449 X_UNLOCK (reslock); 579 X_UNLOCK (reslock);
450 580
451 if (!req) 581 if (!req)
452 return 0; 582 return 0;
453 583
584 X_LOCK (reqlock);
454 --nreqs; 585 --nreqs;
586 X_UNLOCK (reqlock);
455 587
456 if (req->type == EIO_GROUP && req->size) 588 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457 { 589 {
458 req->int1 = 1; /* mark request as delayed */ 590 req->int1 = 1; /* mark request as delayed */
459 continue; 591 continue;
460 } 592 }
461 else 593 else
462 { 594 {
463 int res = eio_finish (req); 595 int res = ETP_FINISH (req);
464 if (res) 596 if (ecb_expect_false (res))
465 return res; 597 return res;
466 } 598 }
467 599
468 if (maxreqs && !--maxreqs) 600 if (ecb_expect_false (maxreqs && !--maxreqs))
469 break; 601 break;
470 602
471 if (max_poll_time) 603 if (maxtime)
472 { 604 {
473 gettimeofday (&tv_now, 0); 605 gettimeofday (&tv_now, 0);
474 606
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 607 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 608 break;
477 } 609 }
478 } 610 }
479 611
480 errno = EAGAIN; 612 errno = EAGAIN;
481 return -1; 613 return -1;
482} 614}
483 615
616static void
617etp_cancel (ETP_REQ *req)
618{
619 req->cancelled = 1;
620
621 eio_grp_cancel (req);
622}
623
624static void
625etp_submit (ETP_REQ *req)
626{
627 req->pri -= ETP_PRI_MIN;
628
629 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
630 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
631
632 if (ecb_expect_false (req->type == EIO_GROUP))
633 {
634 /* I hope this is worth it :/ */
635 X_LOCK (reqlock);
636 ++nreqs;
637 X_UNLOCK (reqlock);
638
639 X_LOCK (reslock);
640
641 ++npending;
642
643 if (!reqq_push (&res_queue, req) && want_poll_cb)
644 want_poll_cb ();
645
646 X_UNLOCK (reslock);
647 }
648 else
649 {
650 X_LOCK (reqlock);
651 ++nreqs;
652 ++nready;
653 reqq_push (&req_queue, req);
654 X_COND_SIGNAL (reqwait);
655 X_UNLOCK (reqlock);
656
657 etp_maybe_start_thread ();
658 }
659}
660
661static void ecb_cold
662etp_set_max_poll_time (double nseconds)
663{
664 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
665 max_poll_time = nseconds * EIO_TICKS;
666 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
667}
668
669static void ecb_cold
670etp_set_max_poll_reqs (unsigned int maxreqs)
671{
672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
673 max_poll_reqs = maxreqs;
674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
675}
676
677static void ecb_cold
678etp_set_max_idle (unsigned int nthreads)
679{
680 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
681 max_idle = nthreads;
682 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
683}
684
685static void ecb_cold
686etp_set_idle_timeout (unsigned int seconds)
687{
688 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
689 idle_timeout = seconds;
690 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
691}
692
693static void ecb_cold
694etp_set_min_parallel (unsigned int nthreads)
695{
696 if (wanted < nthreads)
697 wanted = nthreads;
698}
699
700static void ecb_cold
701etp_set_max_parallel (unsigned int nthreads)
702{
703 if (wanted > nthreads)
704 wanted = nthreads;
705
706 while (started > wanted)
707 etp_end_thread ();
708}
709
710/*****************************************************************************/
711
712static void
713grp_try_feed (eio_req *grp)
714{
715 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
716 {
717 grp->flags &= ~EIO_FLAG_GROUPADD;
718
719 EIO_FEED (grp);
720
721 /* stop if no progress has been made */
722 if (!(grp->flags & EIO_FLAG_GROUPADD))
723 {
724 grp->feed = 0;
725 break;
726 }
727 }
728}
729
730static int
731grp_dec (eio_req *grp)
732{
733 --grp->size;
734
735 /* call feeder, if applicable */
736 grp_try_feed (grp);
737
738 /* finish, if done */
739 if (!grp->size && grp->int1)
740 return eio_finish (grp);
741 else
742 return 0;
743}
744
745static void
746eio_destroy (eio_req *req)
747{
748 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
749 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
750
751 EIO_DESTROY (req);
752}
753
754static int
755eio_finish (eio_req *req)
756{
757 int res = EIO_FINISH (req);
758
759 if (req->grp)
760 {
761 int res2;
762 eio_req *grp = req->grp;
763
764 /* unlink request */
765 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
766 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
767
768 if (grp->grp_first == req)
769 grp->grp_first = req->grp_next;
770
771 res2 = grp_dec (grp);
772
773 if (!res)
774 res = res2;
775 }
776
777 eio_destroy (req);
778
779 return res;
780}
781
782void
783eio_grp_cancel (eio_req *grp)
784{
785 for (grp = grp->grp_first; grp; grp = grp->grp_next)
786 eio_cancel (grp);
787}
788
789void
790eio_cancel (eio_req *req)
791{
792 etp_cancel (req);
793}
794
795void
796eio_submit (eio_req *req)
797{
798 etp_submit (req);
799}
800
801unsigned int
802eio_nreqs (void)
803{
804 return etp_nreqs ();
805}
806
807unsigned int
808eio_nready (void)
809{
810 return etp_nready ();
811}
812
813unsigned int
814eio_npending (void)
815{
816 return etp_npending ();
817}
818
819unsigned int ecb_cold
820eio_nthreads (void)
821{
822 return etp_nthreads ();
823}
824
825void ecb_cold
826eio_set_max_poll_time (double nseconds)
827{
828 etp_set_max_poll_time (nseconds);
829}
830
831void ecb_cold
832eio_set_max_poll_reqs (unsigned int maxreqs)
833{
834 etp_set_max_poll_reqs (maxreqs);
835}
836
837void ecb_cold
838eio_set_max_idle (unsigned int nthreads)
839{
840 etp_set_max_idle (nthreads);
841}
842
843void ecb_cold
844eio_set_idle_timeout (unsigned int seconds)
845{
846 etp_set_idle_timeout (seconds);
847}
848
849void ecb_cold
850eio_set_min_parallel (unsigned int nthreads)
851{
852 etp_set_min_parallel (nthreads);
853}
854
855void ecb_cold
856eio_set_max_parallel (unsigned int nthreads)
857{
858 etp_set_max_parallel (nthreads);
859}
860
861int eio_poll (void)
862{
863 return etp_poll ();
864}
865
484/*****************************************************************************/ 866/*****************************************************************************/
485/* work around various missing functions */ 867/* work around various missing functions */
486 868
487#if !HAVE_PREADWRITE 869#if !HAVE_PREADWRITE
870# undef pread
871# undef pwrite
488# define pread aio_pread 872# define pread eio__pread
489# define pwrite aio_pwrite 873# define pwrite eio__pwrite
490 874
491/* 875static eio_ssize_t
492 * make our pread/pwrite safe against themselves, but not against 876eio__pread (int fd, void *buf, size_t count, off_t offset)
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499{ 877{
500 ssize_t res; 878 eio_ssize_t res;
501 off_t ooffset; 879 off_t ooffset;
502 880
503 X_LOCK (preadwritelock); 881 X_LOCK (preadwritelock);
504 ooffset = lseek (fd, 0, SEEK_CUR); 882 ooffset = lseek (fd, 0, SEEK_CUR);
505 lseek (fd, offset, SEEK_SET); 883 lseek (fd, offset, SEEK_SET);
508 X_UNLOCK (preadwritelock); 886 X_UNLOCK (preadwritelock);
509 887
510 return res; 888 return res;
511} 889}
512 890
513static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 891static eio_ssize_t
892eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514{ 893{
515 ssize_t res; 894 eio_ssize_t res;
516 off_t ooffset; 895 off_t ooffset;
517 896
518 X_LOCK (preadwritelock); 897 X_LOCK (preadwritelock);
519 ooffset = lseek (fd, 0, SEEK_CUR); 898 ooffset = lseek (fd, 0, SEEK_CUR);
520 lseek (fd, offset, SEEK_SET); 899 lseek (fd, offset, SEEK_SET);
521 res = write (fd, buf, count); 900 res = write (fd, buf, count);
522 lseek (fd, offset, SEEK_SET); 901 lseek (fd, ooffset, SEEK_SET);
523 X_UNLOCK (preadwritelock); 902 X_UNLOCK (preadwritelock);
524 903
525 return res; 904 return res;
526} 905}
527#endif 906#endif
528 907
529#ifndef HAVE_FUTIMES 908#ifndef HAVE_UTIMES
530 909
910# undef utimes
531# define utimes(path,times) aio_utimes (path, times) 911# define utimes(path,times) eio__utimes (path, times)
532# define futimes(fd,times) aio_futimes (fd, times)
533 912
913static int
534static int aio_utimes (const char *filename, const struct timeval times[2]) 914eio__utimes (const char *filename, const struct timeval times[2])
535{ 915{
536 if (times) 916 if (times)
537 { 917 {
538 struct utimbuf buf; 918 struct utimbuf buf;
539 919
544 } 924 }
545 else 925 else
546 return utime (filename, 0); 926 return utime (filename, 0);
547} 927}
548 928
929#endif
930
931#ifndef HAVE_FUTIMES
932
933# undef futimes
934# define futimes(fd,times) eio__futimes (fd, times)
935
936static int
549static int aio_futimes (int fd, const struct timeval tv[2]) 937eio__futimes (int fd, const struct timeval tv[2])
550{ 938{
551 errno = ENOSYS; 939 errno = ENOSYS;
552 return -1; 940 return -1;
553} 941}
554 942
555#endif 943#endif
556 944
557#if !HAVE_FDATASYNC 945#if !HAVE_FDATASYNC
946# undef fdatasync
558# define fdatasync fsync 947# define fdatasync(fd) fsync (fd)
559#endif 948#endif
949
950static int
951eio__syncfs (int fd)
952{
953 int res;
954
955#if HAVE_SYS_SYNCFS
956 res = (int)syscall (__NR_syncfs, (int)(fd));
957#else
958 res = -1;
959 errno = ENOSYS;
960#endif
961
962 if (res < 0 && errno == ENOSYS && fd >= 0)
963 sync ();
964
965 return res;
966}
967
968/* sync_file_range always needs emulation */
969static int
970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
971{
972#if HAVE_SYNC_FILE_RANGE
973 int res;
974
975 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
976 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
977 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
978 {
979 flags = 0
980 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
981 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
982 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
983 }
984
985 res = sync_file_range (fd, offset, nbytes, flags);
986
987 if (!res || errno != ENOSYS)
988 return res;
989#endif
990
991 /* even though we could play tricks with the flags, it's better to always
992 * call fdatasync, as that matches the expectation of its users best */
993 return fdatasync (fd);
994}
995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
1005}
560 1006
561#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
1008# undef readahead
562# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563 1010
1011static eio_ssize_t
564static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565{ 1013{
566 size_t todo = count; 1014 size_t todo = count;
567 dBUF; 1015 dBUF;
568 1016
569 while (todo > 0) 1017 while (todo > 0)
573 pread (fd, eio_buf, len, offset); 1021 pread (fd, eio_buf, len, offset);
574 offset += len; 1022 offset += len;
575 todo -= len; 1023 todo -= len;
576 } 1024 }
577 1025
1026 FUBd;
1027
578 errno = 0; 1028 errno = 0;
579 return count; 1029 return count;
580} 1030}
581 1031
582#endif 1032#endif
583 1033
584#if !HAVE_READDIR_R
585# define readdir_r aio_readdir_r
586
587static mutex_t readdirlock = X_MUTEX_INIT;
588
589static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590{
591 X_DIRENT *e;
592 int errorno;
593
594 X_LOCK (readdirlock);
595
596 e = readdir (dirp);
597 errorno = errno;
598
599 if (e)
600 {
601 *res = ent;
602 strcpy (ent->d_name, e->d_name);
603 }
604 else
605 *res = 0;
606
607 X_UNLOCK (readdirlock);
608
609 errno = errorno;
610 return e ? 0 : -1;
611}
612#endif
613
614/* sendfile always needs emulation */ 1034/* sendfile always needs emulation */
1035static eio_ssize_t
615static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1036eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
616{ 1037{
1038 eio_ssize_t written = 0;
617 ssize_t res; 1039 eio_ssize_t res;
618 1040
619 if (!count) 1041 if (!count)
620 return 0; 1042 return 0;
621 1043
1044 for (;;)
1045 {
1046#ifdef __APPLE__
1047# undef HAVE_SENDFILE /* broken, as everything on os x */
1048#endif
622#if HAVE_SENDFILE 1049#if HAVE_SENDFILE
623# if __linux 1050# if __linux
1051 off_t soffset = offset;
624 res = sendfile (ofd, ifd, &offset, count); 1052 res = sendfile (ofd, ifd, &soffset, count);
625 1053
626# elif __freebsd 1054# elif __FreeBSD__
627 /* 1055 /*
628 * Of course, the freebsd sendfile is a dire hack with no thoughts 1056 * Of course, the freebsd sendfile is a dire hack with no thoughts
629 * wasted on making it similar to other I/O functions. 1057 * wasted on making it similar to other I/O functions.
630 */ 1058 */
631 {
632 off_t sbytes; 1059 off_t sbytes;
633 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1060 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634 1061
635 if (res < 0 && sbytes) 1062 #if 0 /* according to the manpage, this is correct, but broken behaviour */
636 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1063 /* freebsd' sendfile will return 0 on success */
1064 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1065 /* not on e.g. EIO or EPIPE - sounds broken */
1066 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637 res = sbytes; 1067 res = sbytes;
638 } 1068 #endif
1069
1070 /* according to source inspection, this is correct, and useful behaviour */
1071 if (sbytes)
1072 res = sbytes;
1073
1074# elif defined (__APPLE__)
1075 off_t sbytes = count;
1076 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1077
1078 /* according to the manpage, sbytes is always valid */
1079 if (sbytes)
1080 res = sbytes;
639 1081
640# elif __hpux 1082# elif __hpux
641 res = sendfile (ofd, ifd, offset, count, 0, 0); 1083 res = sendfile (ofd, ifd, offset, count, 0, 0);
642 1084
643# elif __solaris 1085# elif __solaris
644 {
645 struct sendfilevec vec; 1086 struct sendfilevec vec;
646 size_t sbytes; 1087 size_t sbytes;
647 1088
648 vec.sfv_fd = ifd; 1089 vec.sfv_fd = ifd;
649 vec.sfv_flag = 0; 1090 vec.sfv_flag = 0;
650 vec.sfv_off = offset; 1091 vec.sfv_off = offset;
651 vec.sfv_len = count; 1092 vec.sfv_len = count;
652 1093
653 res = sendfilev (ofd, &vec, 1, &sbytes); 1094 res = sendfilev (ofd, &vec, 1, &sbytes);
654 1095
655 if (res < 0 && sbytes) 1096 if (res < 0 && sbytes)
656 res = sbytes; 1097 res = sbytes;
657 }
658 1098
659# endif 1099# endif
1100
1101#elif defined (_WIN32) && 0
1102 /* does not work, just for documentation of what would need to be done */
1103 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1104 /* libeio guarantees that the file offset does not change, and windows */
1105 /* has no way to get an independent handle to the same file description */
1106 HANDLE h = TO_SOCKET (ifd);
1107 SetFilePointer (h, offset, 0, FILE_BEGIN);
1108 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1109
660#else 1110#else
661 res = -1; 1111 res = -1;
662 errno = ENOSYS; 1112 errno = ENOSYS;
663#endif 1113#endif
664 1114
1115 /* we assume sendfile can copy at least 128mb in one go */
1116 if (res <= 128 * 1024 * 1024)
1117 {
1118 if (res > 0)
1119 written += res;
1120
1121 if (written)
1122 return written;
1123
1124 break;
1125 }
1126 else
1127 {
1128 /* if we requested more, then probably the kernel was lazy */
1129 written += res;
1130 offset += res;
1131 count -= res;
1132
1133 if (!count)
1134 return written;
1135 }
1136 }
1137
665 if (res < 0 1138 if (res < 0
666 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1139 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1140 /* BSDs */
1141#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1142 || errno == ENOTSUP
1143#endif
1144#ifdef EOPNOTSUPP /* windows */
1145 || errno == EOPNOTSUPP /* BSDs */
1146#endif
667#if __solaris 1147#if __solaris
668 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1148 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
669#endif 1149#endif
670 ) 1150 )
671 ) 1151 )
675 1155
676 res = 0; 1156 res = 0;
677 1157
678 while (count) 1158 while (count)
679 { 1159 {
680 ssize_t cnt; 1160 eio_ssize_t cnt;
681 1161
682 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1162 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
683 1163
684 if (cnt <= 0) 1164 if (cnt <= 0)
685 { 1165 {
697 1177
698 offset += cnt; 1178 offset += cnt;
699 res += cnt; 1179 res += cnt;
700 count -= cnt; 1180 count -= cnt;
701 } 1181 }
1182
1183 FUBd;
702 } 1184 }
703 1185
704 return res; 1186 return res;
705} 1187}
706 1188
1189#ifdef PAGESIZE
1190# define eio_pagesize() PAGESIZE
1191#else
1192static intptr_t
1193eio_pagesize (void)
1194{
1195 static intptr_t page;
1196
1197 if (!page)
1198 page = sysconf (_SC_PAGESIZE);
1199
1200 return page;
1201}
1202#endif
1203
1204static void
1205eio_page_align (void **addr, size_t *length)
1206{
1207 intptr_t mask = eio_pagesize () - 1;
1208
1209 /* round down addr */
1210 intptr_t adj = mask & (intptr_t)*addr;
1211
1212 *addr = (void *)((intptr_t)*addr - adj);
1213 *length += adj;
1214
1215 /* round up length */
1216 *length = (*length + mask) & ~mask;
1217}
1218
1219#if !_POSIX_MEMLOCK
1220# define eio__mlockall(a) EIO_ENOSYS ()
1221#else
1222
1223static int
1224eio__mlockall (int flags)
1225{
1226 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1227 extern int mallopt (int, int);
1228 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1229 #endif
1230
1231 if (EIO_MCL_CURRENT != MCL_CURRENT
1232 || EIO_MCL_FUTURE != MCL_FUTURE)
1233 {
1234 flags = 0
1235 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1236 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1237 }
1238
1239 return mlockall (flags);
1240}
1241#endif
1242
1243#if !_POSIX_MEMLOCK_RANGE
1244# define eio__mlock(a,b) EIO_ENOSYS ()
1245#else
1246
1247static int
1248eio__mlock (void *addr, size_t length)
1249{
1250 eio_page_align (&addr, &length);
1251
1252 return mlock (addr, length);
1253}
1254
1255#endif
1256
1257#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1258# define eio__msync(a,b,c) EIO_ENOSYS ()
1259#else
1260
1261static int
1262eio__msync (void *mem, size_t len, int flags)
1263{
1264 eio_page_align (&mem, &len);
1265
1266 if (EIO_MS_ASYNC != MS_SYNC
1267 || EIO_MS_INVALIDATE != MS_INVALIDATE
1268 || EIO_MS_SYNC != MS_SYNC)
1269 {
1270 flags = 0
1271 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1272 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1273 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1274 }
1275
1276 return msync (mem, len, flags);
1277}
1278
1279#endif
1280
1281static int
1282eio__mtouch (eio_req *req)
1283{
1284 void *mem = req->ptr2;
1285 size_t len = req->size;
1286 int flags = req->int1;
1287
1288 eio_page_align (&mem, &len);
1289
1290 {
1291 intptr_t addr = (intptr_t)mem;
1292 intptr_t end = addr + len;
1293 intptr_t page = eio_pagesize ();
1294
1295 if (addr < end)
1296 if (flags & EIO_MT_MODIFY) /* modify */
1297 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1298 else
1299 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1300 }
1301
1302 return 0;
1303}
1304
1305/*****************************************************************************/
1306/* requests implemented outside eio_execute, because they are so large */
1307
1308static void
1309eio__realpath (eio_req *req, etp_worker *self)
1310{
1311 char *rel = req->ptr1;
1312 char *res;
1313 char *tmp1, *tmp2;
1314#if SYMLOOP_MAX > 32
1315 int symlinks = SYMLOOP_MAX;
1316#else
1317 int symlinks = 32;
1318#endif
1319
1320 req->result = -1;
1321
1322 errno = EINVAL;
1323 if (!rel)
1324 return;
1325
1326 errno = ENOENT;
1327 if (!*rel)
1328 return;
1329
1330 if (!req->ptr2)
1331 {
1332 X_LOCK (wrklock);
1333 req->flags |= EIO_FLAG_PTR2_FREE;
1334 X_UNLOCK (wrklock);
1335 req->ptr2 = malloc (PATH_MAX * 3);
1336
1337 errno = ENOMEM;
1338 if (!req->ptr2)
1339 return;
1340 }
1341
1342 res = req->ptr2;
1343 tmp1 = res + PATH_MAX;
1344 tmp2 = tmp1 + PATH_MAX;
1345
1346#if 0 /* disabled, the musl way to do things is just too racy */
1347#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1348 /* on linux we may be able to ask the kernel */
1349 {
1350 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1351
1352 if (fd >= 0)
1353 {
1354 sprintf (tmp1, "/proc/self/fd/%d", fd);
1355 req->result = readlink (tmp1, res, PATH_MAX);
1356 close (fd);
1357
1358 /* here we should probably stat the open file and the disk file, to make sure they still match */
1359
1360 if (req->result > 0)
1361 goto done;
1362 }
1363 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1364 return;
1365 }
1366#endif
1367#endif
1368
1369 if (*rel != '/')
1370 {
1371 if (!getcwd (res, PATH_MAX))
1372 return;
1373
1374 if (res [1]) /* only use if not / */
1375 res += strlen (res);
1376 }
1377
1378 while (*rel)
1379 {
1380 eio_ssize_t len, linklen;
1381 char *beg = rel;
1382
1383 while (*rel && *rel != '/')
1384 ++rel;
1385
1386 len = rel - beg;
1387
1388 if (!len) /* skip slashes */
1389 {
1390 ++rel;
1391 continue;
1392 }
1393
1394 if (beg [0] == '.')
1395 {
1396 if (len == 1)
1397 continue; /* . - nop */
1398
1399 if (beg [1] == '.' && len == 2)
1400 {
1401 /* .. - back up one component, if possible */
1402
1403 while (res != req->ptr2)
1404 if (*--res == '/')
1405 break;
1406
1407 continue;
1408 }
1409 }
1410
1411 errno = ENAMETOOLONG;
1412 if (res + 1 + len + 1 >= tmp1)
1413 return;
1414
1415 /* copy one component */
1416 *res = '/';
1417 memcpy (res + 1, beg, len);
1418
1419 /* zero-terminate, for readlink */
1420 res [len + 1] = 0;
1421
1422 /* now check if it's a symlink */
1423 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1424
1425 if (linklen < 0)
1426 {
1427 if (errno != EINVAL)
1428 return;
1429
1430 /* it's a normal directory. hopefully */
1431 res += len + 1;
1432 }
1433 else
1434 {
1435 /* yay, it was a symlink - build new path in tmp2 */
1436 int rellen = strlen (rel);
1437
1438 errno = ENAMETOOLONG;
1439 if (linklen + 1 + rellen >= PATH_MAX)
1440 return;
1441
1442 errno = ELOOP;
1443 if (!--symlinks)
1444 return;
1445
1446 if (*tmp1 == '/')
1447 res = req->ptr2; /* symlink resolves to an absolute path */
1448
1449 /* we need to be careful, as rel might point into tmp2 already */
1450 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1451 tmp2 [linklen] = '/';
1452 memcpy (tmp2, tmp1, linklen);
1453
1454 rel = tmp2;
1455 }
1456 }
1457
1458 /* special case for the lone root path */
1459 if (res == req->ptr2)
1460 *res++ = '/';
1461
1462 req->result = res - (char *)req->ptr2;
1463
1464done:
1465 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1466}
1467
1468static signed char
1469eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1470{
1471 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1472 : a->inode < b->inode ? -1
1473 : a->inode > b->inode ? 1
1474 : 0;
1475}
1476
1477#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1478
1479#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1480#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1481
1482static void
1483eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1484{
1485 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1486 unsigned char *bit = bits;
1487
1488 assert (CHAR_BIT == 8);
1489 assert (sizeof (eio_dirent) * 8 < 256);
1490 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1491 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1492
1493 if (size <= EIO_SORT_FAST)
1494 return;
1495
1496 /* first prepare an array of bits to test in our radix sort */
1497 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1498 /* inode_bits must contain all inodes ORed together */
1499 /* which is used to skip bits that are 0 everywhere, which is very common */
1500 {
1501 eio_ino_t endianness;
1502 int i, j;
1503
1504 /* we store the byte offset of byte n into byte n of "endianness" */
1505 for (i = 0; i < sizeof (eio_ino_t); ++i)
1506 ((unsigned char *)&endianness)[i] = i;
1507
1508 *bit++ = 0;
1509
1510 for (i = 0; i < sizeof (eio_ino_t); ++i)
1511 {
1512 /* shifting off the byte offsets out of "endianness" */
1513 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1514 endianness >>= 8;
1515
1516 for (j = 0; j < 8; ++j)
1517 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1518 *bit++ = offs + j;
1519 }
1520
1521 for (j = 0; j < 8; ++j)
1522 if (score_bits & (1 << j))
1523 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1524 }
1525
1526 /* now actually do the sorting (a variant of MSD radix sort) */
1527 {
1528 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1529 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1530 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1531 int stk_idx = 0;
1532
1533 base_stk [stk_idx] = dents;
1534 end_stk [stk_idx] = dents + size;
1535 bit_stk [stk_idx] = bit - 1;
1536
1537 do
1538 {
1539 base = base_stk [stk_idx];
1540 end = end_stk [stk_idx];
1541 bit = bit_stk [stk_idx];
1542
1543 for (;;)
1544 {
1545 unsigned char O = *bit >> 3;
1546 unsigned char M = 1 << (*bit & 7);
1547
1548 eio_dirent *a = base;
1549 eio_dirent *b = end;
1550
1551 if (b - a < EIO_SORT_CUTOFF)
1552 break;
1553
1554 /* now bit-partition the array on the bit */
1555 /* this ugly asymmetric loop seems to perform much better than typical */
1556 /* partition algos found in the literature */
1557 do
1558 if (!(((unsigned char *)a)[O] & M))
1559 ++a;
1560 else if (!(((unsigned char *)--b)[O] & M))
1561 {
1562 eio_dirent tmp = *a; *a = *b; *b = tmp;
1563 ++a;
1564 }
1565 while (b > a);
1566
1567 /* next bit, or stop, if no bits left in this path */
1568 if (!*--bit)
1569 break;
1570
1571 base_stk [stk_idx] = a;
1572 end_stk [stk_idx] = end;
1573 bit_stk [stk_idx] = bit;
1574 ++stk_idx;
1575
1576 end = a;
1577 }
1578 }
1579 while (stk_idx--);
1580 }
1581}
1582
1583static void
1584eio_dent_insertion_sort (eio_dirent *dents, int size)
1585{
1586 /* first move the smallest element to the front, to act as a sentinel */
1587 {
1588 int i;
1589 eio_dirent *min = dents;
1590
1591 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1592 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1593 if (EIO_DENT_CMP (dents [i], <, *min))
1594 min = &dents [i];
1595
1596 /* swap elements 0 and j (minimum) */
1597 {
1598 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1599 }
1600 }
1601
1602 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1603 {
1604 eio_dirent *i, *j;
1605
1606 for (i = dents + 1; i < dents + size; ++i)
1607 {
1608 eio_dirent value = *i;
1609
1610 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1611 j [1] = j [0];
1612
1613 j [1] = value;
1614 }
1615 }
1616}
1617
1618static void
1619eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1620{
1621 if (size <= 1)
1622 return; /* our insertion sort relies on size > 0 */
1623
1624 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1625 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1626 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1627
1628 /* use an insertion sort at the end, or for small arrays, */
1629 /* as insertion sort is more efficient for small partitions */
1630 eio_dent_insertion_sort (dents, size);
1631}
1632
707/* read a full directory */ 1633/* read a full directory */
1634static void
708static void scandir_ (eio_req *req, worker *self) 1635eio__scandir (eio_req *req, etp_worker *self)
709{ 1636{
1637 char *name, *names;
1638 int namesalloc = 4096 - sizeof (void *) * 4;
1639 int namesoffs = 0;
1640 int flags = req->int1;
1641 eio_dirent *dents = 0;
1642 int dentalloc = 128;
1643 int dentoffs = 0;
1644 eio_ino_t inode_bits = 0;
1645#ifdef _WIN32
1646 HANDLE dirp;
1647 WIN32_FIND_DATA entp;
1648#else
710 DIR *dirp; 1649 DIR *dirp;
711 union
712 {
713 EIO_STRUCT_DIRENT d;
714 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715 } *u;
716 EIO_STRUCT_DIRENT *entp; 1650 EIO_STRUCT_DIRENT *entp;
717 char *name, *names; 1651#endif
718 int memlen = 4096;
719 int memofs = 0;
720 int res = 0;
721 1652
722 X_LOCK (wrklock); 1653 req->result = -1;
1654
1655 if (!(flags & EIO_READDIR_DENTS))
1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1657
1658#ifdef _WIN32
1659 {
1660 int len = strlen ((const char *)req->ptr1);
1661 char *path = malloc (MAX_PATH);
1662 const char *fmt;
1663
1664 if (!len)
1665 fmt = "./*";
1666 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1667 fmt = "%s*";
1668 else
1669 fmt = "%s/*";
1670
1671 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1672 dirp = FindFirstFile (path, &entp);
1673 free (path);
1674
1675 if (dirp == INVALID_HANDLE_VALUE)
1676 {
1677 dirp = 0;
1678
1679 switch (GetLastError ())
1680 {
1681 case ERROR_FILE_NOT_FOUND:
1682 req->result = 0;
1683 break;
1684
1685 case ERROR_INVALID_NAME:
1686 case ERROR_PATH_NOT_FOUND:
1687 case ERROR_NO_MORE_FILES:
1688 errno = ENOENT;
1689 break;
1690
1691 case ERROR_NOT_ENOUGH_MEMORY:
1692 errno = ENOMEM;
1693 break;
1694
1695 default:
1696 errno = EINVAL;
1697 break;
1698 }
1699 }
1700 }
1701#else
723 self->dirp = dirp = opendir (req->ptr1); 1702 dirp = opendir (req->ptr1);
724 self->dbuf = u = malloc (sizeof (*u)); 1703#endif
1704
725 req->flags |= EIO_FLAG_PTR2_FREE; 1705 if (req->flags & EIO_FLAG_PTR1_FREE)
1706 free (req->ptr1);
1707
1708 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1709 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726 req->ptr2 = names = malloc (memlen); 1710 req->ptr2 = names = malloc (namesalloc);
727 X_UNLOCK (wrklock);
728 1711
729 if (dirp && u && names) 1712 if (dirp && names && (!flags || dents))
730 for (;;) 1713 for (;;)
731 { 1714 {
1715 int done;
1716
1717#ifdef _WIN32
1718 done = !dirp;
1719#else
732 errno = 0; 1720 errno = 0;
733 readdir_r (dirp, &u->d, &entp); 1721 entp = readdir (dirp);
1722 done = !entp;
1723#endif
734 1724
735 if (!entp) 1725 if (done)
1726 {
1727#ifndef _WIN32
1728 int old_errno = errno;
1729 closedir (dirp);
1730 errno = old_errno;
1731
1732 if (errno)
1733 break;
1734#endif
1735
1736 /* sort etc. */
1737 req->int1 = flags;
1738 req->result = dentoffs;
1739
1740 if (flags & EIO_READDIR_STAT_ORDER)
1741 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1742 else if (flags & EIO_READDIR_DIRS_FIRST)
1743 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1744 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1745 else
1746 {
1747 /* in this case, all is known, and we just put dirs first and sort them */
1748 eio_dirent *oth = dents + dentoffs;
1749 eio_dirent *dir = dents;
1750
1751 /* now partition dirs to the front, and non-dirs to the back */
1752 /* by walking from both sides and swapping if necessary */
1753 while (oth > dir)
1754 {
1755 if (dir->type == EIO_DT_DIR)
1756 ++dir;
1757 else if ((--oth)->type == EIO_DT_DIR)
1758 {
1759 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1760
1761 ++dir;
1762 }
1763 }
1764
1765 /* now sort the dirs only (dirs all have the same score) */
1766 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1767 }
1768
736 break; 1769 break;
1770 }
737 1771
738 name = entp->d_name; 1772 /* now add the entry to our list(s) */
1773 name = D_NAME (entp);
739 1774
1775 /* skip . and .. entries */
740 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1776 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741 { 1777 {
742 int len = strlen (name) + 1; 1778 int len = D_NAMLEN (entp) + 1;
743 1779
744 res++; 1780 while (ecb_expect_false (namesoffs + len > namesalloc))
745
746 while (memofs + len > memlen)
747 { 1781 {
748 memlen *= 2; 1782 namesalloc *= 2;
749 X_LOCK (wrklock);
750 req->ptr2 = names = realloc (names, memlen); 1783 req->ptr2 = names = realloc (names, namesalloc);
751 X_UNLOCK (wrklock);
752 1784
753 if (!names) 1785 if (!names)
754 break; 1786 break;
755 } 1787 }
756 1788
757 memcpy (names + memofs, name, len); 1789 memcpy (names + namesoffs, name, len);
1790
1791 if (dents)
1792 {
1793 struct eio_dirent *ent;
1794
1795 if (ecb_expect_false (dentoffs == dentalloc))
1796 {
1797 dentalloc *= 2;
1798 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1799
1800 if (!dents)
1801 break;
1802 }
1803
1804 ent = dents + dentoffs;
1805
1806 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1807 ent->namelen = len - 1;
1808 ent->inode = D_INO (entp);
1809
1810 inode_bits |= ent->inode;
1811
1812 switch (D_TYPE (entp))
1813 {
1814 default:
1815 ent->type = EIO_DT_UNKNOWN;
1816 flags |= EIO_READDIR_FOUND_UNKNOWN;
1817 break;
1818
1819 #ifdef DT_FIFO
1820 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1821 #endif
1822 #ifdef DT_CHR
1823 case DT_CHR: ent->type = EIO_DT_CHR; break;
1824 #endif
1825 #ifdef DT_MPC
1826 case DT_MPC: ent->type = EIO_DT_MPC; break;
1827 #endif
1828 #ifdef DT_DIR
1829 case DT_DIR: ent->type = EIO_DT_DIR; break;
1830 #endif
1831 #ifdef DT_NAM
1832 case DT_NAM: ent->type = EIO_DT_NAM; break;
1833 #endif
1834 #ifdef DT_BLK
1835 case DT_BLK: ent->type = EIO_DT_BLK; break;
1836 #endif
1837 #ifdef DT_MPB
1838 case DT_MPB: ent->type = EIO_DT_MPB; break;
1839 #endif
1840 #ifdef DT_REG
1841 case DT_REG: ent->type = EIO_DT_REG; break;
1842 #endif
1843 #ifdef DT_NWK
1844 case DT_NWK: ent->type = EIO_DT_NWK; break;
1845 #endif
1846 #ifdef DT_CMP
1847 case DT_CMP: ent->type = EIO_DT_CMP; break;
1848 #endif
1849 #ifdef DT_LNK
1850 case DT_LNK: ent->type = EIO_DT_LNK; break;
1851 #endif
1852 #ifdef DT_SOCK
1853 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1854 #endif
1855 #ifdef DT_DOOR
1856 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1857 #endif
1858 #ifdef DT_WHT
1859 case DT_WHT: ent->type = EIO_DT_WHT; break;
1860 #endif
1861 }
1862
1863 ent->score = 7;
1864
1865 if (flags & EIO_READDIR_DIRS_FIRST)
1866 {
1867 if (ent->type == EIO_DT_UNKNOWN)
1868 {
1869 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1870 ent->score = 1;
1871 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1872 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1873 }
1874 else if (ent->type == EIO_DT_DIR)
1875 ent->score = 0;
1876 }
1877 }
1878
758 memofs += len; 1879 namesoffs += len;
1880 ++dentoffs;
759 } 1881 }
1882
1883 if (EIO_CANCELLED (req))
1884 {
1885 errno = ECANCELED;
1886 break;
1887 }
1888
1889#ifdef _WIN32
1890 if (!FindNextFile (dirp, &entp))
1891 {
1892 FindClose (dirp);
1893 dirp = 0;
1894 }
1895#endif
760 } 1896 }
761
762 if (errno)
763 res = -1;
764
765 req->result = res;
766} 1897}
767 1898
768/*****************************************************************************/ 1899/*****************************************************************************/
769 1900
1901#define ALLOC(len) \
1902 if (!req->ptr2) \
1903 { \
1904 X_LOCK (wrklock); \
1905 req->flags |= EIO_FLAG_PTR2_FREE; \
1906 X_UNLOCK (wrklock); \
1907 req->ptr2 = malloc (len); \
1908 if (!req->ptr2) \
1909 { \
1910 errno = ENOMEM; \
1911 req->result = -1; \
1912 break; \
1913 } \
1914 }
1915
770X_THREAD_PROC (eio_proc) 1916X_THREAD_PROC (etp_proc)
771{ 1917{
772 eio_req *req; 1918 ETP_REQ *req;
773 struct timespec ts; 1919 struct timespec ts;
774 worker *self = (worker *)thr_arg; 1920 etp_worker *self = (etp_worker *)thr_arg;
775 1921
1922#if HAVE_PRCTL_SET_NAME
1923 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1924#endif
1925
776 /* try to distribute timeouts somewhat randomly */ 1926 /* try to distribute timeouts somewhat evenly */
777 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1927 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778 1928
779 for (;;) 1929 for (;;)
780 { 1930 {
781 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1931 ts.tv_sec = 0;
782 1932
783 X_LOCK (reqlock); 1933 X_LOCK (reqlock);
784 1934
785 for (;;) 1935 for (;;)
786 { 1936 {
787 self->req = req = reqq_shift (&req_queue); 1937 self->req = req = reqq_shift (&req_queue);
788 1938
789 if (req) 1939 if (req)
790 break; 1940 break;
791 1941
1942 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1943 {
1944 X_UNLOCK (reqlock);
1945 X_LOCK (wrklock);
1946 --started;
1947 X_UNLOCK (wrklock);
1948 goto quit;
1949 }
1950
792 ++idle; 1951 ++idle;
793 1952
794 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1953 if (idle <= max_idle)
1954 /* we are allowed to idle, so do so without any timeout */
1955 X_COND_WAIT (reqwait, reqlock);
1956 else
795 { 1957 {
796 if (idle > max_idle) 1958 /* initialise timeout once */
797 { 1959 if (!ts.tv_sec)
798 --idle; 1960 ts.tv_sec = time (0) + idle_timeout;
799 X_UNLOCK (reqlock);
800 X_LOCK (wrklock);
801 --started;
802 X_UNLOCK (wrklock);
803 goto quit;
804 }
805 1961
806 /* we are allowed to idle, so do so without any timeout */
807 X_COND_WAIT (reqwait, reqlock); 1962 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1963 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
809 } 1964 }
810 1965
811 --idle; 1966 --idle;
812 } 1967 }
813 1968
814 --nready; 1969 --nready;
815 1970
816 X_UNLOCK (reqlock); 1971 X_UNLOCK (reqlock);
817 1972
818 errno = 0; /* strictly unnecessary */
819
820 if (!EIO_CANCELLED (req))
821 switch (req->type) 1973 if (req->type < 0)
822 {
823 case EIO_READ: req->result = req->offs >= 0
824 ? pread (req->int1, req->ptr2, req->size, req->offs)
825 : read (req->int1, req->ptr2, req->size); break;
826 case EIO_WRITE: req->result = req->offs >= 0
827 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
828 : write (req->int1, req->ptr2, req->size); break;
829
830 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845 case EIO_CLOSE: req->result = close (req->int1); break;
846 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856 case EIO_SYNC: req->result = 0; sync (); break;
857 case EIO_FSYNC: req->result = fsync (req->int1); break;
858 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860 case EIO_READDIR: scandir_ (req, self); break;
861
862 case EIO_BUSY:
863#ifdef _WIN32
864 Sleep (req->nv1 * 1000.);
865#else
866 {
867 struct timeval tv;
868
869 tv.tv_sec = req->nv1;
870 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872 req->result = select (0, 0, 0, 0, &tv);
873 }
874#endif
875 break;
876
877 case EIO_UTIME:
878 case EIO_FUTIME:
879 {
880 struct timeval tv[2];
881 struct timeval *times;
882
883 if (req->nv1 != -1. || req->nv2 != -1.)
884 {
885 tv[0].tv_sec = req->nv1;
886 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887 tv[1].tv_sec = req->nv2;
888 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890 times = tv;
891 }
892 else
893 times = 0;
894
895
896 req->result = req->type == EIO_FUTIME
897 ? futimes (req->int1, times)
898 : utimes (req->ptr1, times);
899 }
900
901 case EIO_GROUP:
902 case EIO_NOP:
903 break;
904
905 case EIO_QUIT:
906 goto quit; 1974 goto quit;
907 1975
908 default: 1976 ETP_EXECUTE (self, req);
909 req->result = -1;
910 break;
911 }
912
913 req->errorno = errno;
914 1977
915 X_LOCK (reslock); 1978 X_LOCK (reslock);
916 1979
917 ++npending; 1980 ++npending;
918 1981
919 if (!reqq_push (&res_queue, req) && want_poll_cb) 1982 if (!reqq_push (&res_queue, req) && want_poll_cb)
920 want_poll_cb (); 1983 want_poll_cb ();
921 1984
922 self->req = 0; 1985 self->req = 0;
923 worker_clear (self); 1986 etp_worker_clear (self);
924 1987
925 X_UNLOCK (reslock); 1988 X_UNLOCK (reslock);
926 } 1989 }
927 1990
928quit: 1991quit:
929 X_LOCK (wrklock); 1992 X_LOCK (wrklock);
930 worker_free (self); 1993 etp_worker_free (self);
931 X_UNLOCK (wrklock); 1994 X_UNLOCK (wrklock);
932 1995
933 return 0; 1996 return 0;
934} 1997}
935 1998
936/*****************************************************************************/ 1999/*****************************************************************************/
937 2000
938static void eio_atfork_prepare (void) 2001int ecb_cold
2002eio_init (void (*want_poll)(void), void (*done_poll)(void))
939{ 2003{
940 X_LOCK (wrklock);
941 X_LOCK (reqlock);
942 X_LOCK (reslock);
943#if !HAVE_PREADWRITE 2004#if !HAVE_PREADWRITE
944 X_LOCK (preadwritelock);
945#endif
946#if !HAVE_READDIR_R
947 X_LOCK (readdirlock);
948#endif
949}
950
951static void eio_atfork_parent (void)
952{
953#if !HAVE_READDIR_R
954 X_UNLOCK (readdirlock);
955#endif
956#if !HAVE_PREADWRITE
957 X_UNLOCK (preadwritelock); 2005 X_MUTEX_CREATE (preadwritelock);
958#endif 2006#endif
959 X_UNLOCK (reslock);
960 X_UNLOCK (reqlock);
961 X_UNLOCK (wrklock);
962}
963 2007
964static void eio_atfork_child (void) 2008 return etp_init (want_poll, done_poll);
965{ 2009}
966 eio_req *prv;
967 2010
968 while (prv = reqq_shift (&req_queue)) 2011ecb_inline void
969 eio_destroy (prv); 2012eio_api_destroy (eio_req *req)
2013{
2014 free (req);
2015}
970 2016
971 while (prv = reqq_shift (&res_queue)) 2017#define REQ(rtype) \
972 eio_destroy (prv); 2018 eio_req *req; \
2019 \
2020 req = (eio_req *)calloc (1, sizeof *req); \
2021 if (!req) \
2022 return 0; \
2023 \
2024 req->type = rtype; \
2025 req->pri = pri; \
2026 req->finish = cb; \
2027 req->data = data; \
2028 req->destroy = eio_api_destroy;
973 2029
974 while (wrk_first.next != &wrk_first) 2030#define SEND eio_submit (req); return req
2031
2032#define PATH \
2033 req->flags |= EIO_FLAG_PTR1_FREE; \
2034 req->ptr1 = strdup (path); \
2035 if (!req->ptr1) \
2036 { \
2037 eio_api_destroy (req); \
2038 return 0; \
2039 }
2040
2041static void
2042eio_execute (etp_worker *self, eio_req *req)
2043{
2044 if (ecb_expect_false (EIO_CANCELLED (req)))
975 { 2045 {
976 worker *wrk = wrk_first.next; 2046 req->result = -1;
977 2047 req->errorno = ECANCELED;
978 if (wrk->req) 2048 return;
979 eio_destroy (wrk->req);
980
981 worker_clear (wrk);
982 worker_free (wrk);
983 } 2049 }
984 2050
985 started = 0; 2051 switch (req->type)
986 idle = 0; 2052 {
987 nreqs = 0; 2053 case EIO_READ: ALLOC (req->size);
988 nready = 0; 2054 req->result = req->offs >= 0
989 npending = 0; 2055 ? pread (req->int1, req->ptr2, req->size, req->offs)
2056 : read (req->int1, req->ptr2, req->size); break;
2057 case EIO_WRITE: req->result = req->offs >= 0
2058 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2059 : write (req->int1, req->ptr2, req->size); break;
990 2060
991 eio_atfork_parent (); 2061 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
992} 2062 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
993 2063
994int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2064 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
995{ 2065 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
996 want_poll_cb = want_poll; 2066 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
997 done_poll_cb = done_poll; 2067 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2068 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2069 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
998 2070
2071 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2072 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2073 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2074 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2075
2076 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2077 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2078 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2079 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2080 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2081 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2082
2083 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2084 case EIO_CLOSE: req->result = close (req->int1); break;
2085 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2086 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2087 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2088 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2089 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2090 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2091 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2092 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2093
2094 case EIO_REALPATH: eio__realpath (req, self); break;
2095
2096 case EIO_READLINK: ALLOC (PATH_MAX);
2097 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2098
2099 case EIO_SYNC: req->result = 0; sync (); break;
2100 case EIO_FSYNC: req->result = fsync (req->int1); break;
2101 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2102 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2103 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2104 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2105 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2106 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2107 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2108 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2109
2110 case EIO_READDIR: eio__scandir (req, self); break;
2111
2112 case EIO_BUSY:
999#ifdef _WIN32 2113#ifdef _WIN32
1000 X_MUTEX_CHECK (wrklock); 2114 Sleep (req->nv1 * 1e3);
1001 X_MUTEX_CHECK (reslock); 2115#else
1002 X_MUTEX_CHECK (reqlock); 2116 {
1003 X_MUTEX_CHECK (reqwait); 2117 struct timeval tv;
1004 X_MUTEX_CHECK (preadwritelock);
1005 X_MUTEX_CHECK (readdirlock);
1006 2118
1007 X_COND_CHECK (reqwait); 2119 tv.tv_sec = req->nv1;
1008#endif 2120 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1009 2121
1010 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2122 req->result = select (0, 0, 0, 0, &tv);
1011} 2123 }
2124#endif
2125 break;
1012 2126
1013#if 0 2127 case EIO_UTIME:
2128 case EIO_FUTIME:
2129 {
2130 struct timeval tv[2];
2131 struct timeval *times;
1014 2132
2133 if (req->nv1 != -1. || req->nv2 != -1.)
2134 {
2135 tv[0].tv_sec = req->nv1;
2136 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2137 tv[1].tv_sec = req->nv2;
2138 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2139
2140 times = tv;
2141 }
2142 else
2143 times = 0;
2144
2145 req->result = req->type == EIO_FUTIME
2146 ? futimes (req->int1, times)
2147 : utimes (req->ptr1, times);
2148 }
2149 break;
2150
2151 case EIO_GROUP:
2152 abort (); /* handled in eio_request */
2153
2154 case EIO_NOP:
2155 req->result = 0;
2156 break;
2157
2158 case EIO_CUSTOM:
2159 req->feed (req);
2160 break;
2161
2162 default:
2163 errno = ENOSYS;
2164 req->result = -1;
2165 break;
2166 }
2167
2168 req->errorno = errno;
2169}
2170
2171#ifndef EIO_NO_WRAPPERS
2172
2173eio_req *eio_nop (int pri, eio_cb cb, void *data)
2174{
2175 REQ (EIO_NOP); SEND;
2176}
2177
2178eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2179{
2180 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2181}
2182
1015eio_req *eio_fsync (int fd, eio_cb cb); 2183eio_req *eio_sync (int pri, eio_cb cb, void *data)
2184{
2185 REQ (EIO_SYNC); SEND;
2186}
2187
2188eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2189{
2190 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2191}
2192
2193eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2194{
2195 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2196}
2197
1016eio_req *eio_fdatasync (int fd, eio_cb cb); 2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017eio_req *eio_dupclose (int fd, eio_cb cb); 2199{
2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2201}
2202
2203eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2204{
2205 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2206}
2207
2208eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2209{
2210 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2211}
2212
2213eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2214{
2215 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2216}
2217
2218eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2219{
2220 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2221}
2222
2223eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2224{
2225 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2226}
2227
2228eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2231}
2232
2233eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2236}
2237
1018eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2238eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2239{
1020eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2240 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2241}
2242
2243eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2246}
2247
2248eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2249{
2250 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2251}
2252
2253eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2254{
2255 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2256}
2257
2258eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2259{
2260 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2261}
2262
1022eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2263eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2264{
2265 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2266}
2267
1023eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2268eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2269{
2270 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2271}
2272
1024eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2273eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2274{
2275 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2276}
2277
1025eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2278eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2279{
2280 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2281}
2282
1026eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2283eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2284{
2285 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2286}
2287
1027eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2288eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2289{
2290 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2291}
2292
1028eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2293eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2294{
1030eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2295 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2296}
2297
1032eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2298eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2299{
2300 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2301}
2302
1033eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2303eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2304{
2305 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2306}
2307
2308eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2309{
2310 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2311}
2312
1034eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2313eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2314{
2315 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2316}
2317
1035eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2318eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1036eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2319{
2320 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2321}
2322
2323static eio_req *
2324eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2325{
2326 REQ (type); PATH; SEND;
2327}
2328
2329eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2330{
2331 return eio__1path (EIO_READLINK, path, pri, cb, data);
2332}
2333
2334eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2335{
2336 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2337}
2338
2339eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2340{
2341 return eio__1path (EIO_STAT, path, pri, cb, data);
2342}
2343
2344eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2345{
2346 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2347}
2348
2349eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2350{
2351 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2352}
2353
1037eio_req *eio_unlink (const char *path, eio_cb cb); 2354eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2355{
2356 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2357}
2358
1038eio_req *eio_rmdir (const char *path, eio_cb cb); 2359eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2360{
2361 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2362}
2363
2364eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2365{
2366 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2367}
2368
1040eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2369eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2370{
1042eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2371 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2372}
2373
2374static eio_req *
2375eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2376{
2377 REQ (type); PATH;
2378
2379 req->flags |= EIO_FLAG_PTR2_FREE;
2380 req->ptr2 = strdup (new_path);
2381 if (!req->ptr2)
2382 {
2383 eio_api_destroy (req);
2384 return 0;
2385 }
2386
2387 SEND;
2388}
2389
2390eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2391{
2392 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2393}
2394
2395eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2396{
2397 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2398}
2399
2400eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2401{
2402 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2403}
2404
2405eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2406{
2407 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2408}
2409
2410#endif
2411
2412eio_req *eio_grp (eio_cb cb, void *data)
2413{
2414 const int pri = EIO_PRI_MAX;
2415
2416 REQ (EIO_GROUP); SEND;
2417}
2418
2419#undef REQ
2420#undef PATH
2421#undef SEND
2422
2423/*****************************************************************************/
2424/* grp functions */
2425
1043void 2426void
1044aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045 PROTOTYPE: $$$;$
1046 PPCODE:
1047{
1048 dREQ;
1049
1050 req->type = EIO_OPEN;
1051 req->sv1 = newSVsv (pathname);
1052 req->ptr1 = SvPVbyte_nolen (req->sv1);
1053 req->int1 = flags;
1054 req->int2 = mode;
1055
1056 EIO_SEND;
1057}
1058
1059void
1060aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061 PROTOTYPE: $;$
1062 ALIAS:
1063 aio_fsync = EIO_FSYNC
1064 aio_fdatasync = EIO_FDATASYNC
1065 PPCODE:
1066{
1067 dREQ;
1068
1069 req->type = ix;
1070 req->sv1 = newSVsv (fh);
1071 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073 EIO_SEND (req);
1074}
1075
1076void
1077aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078 PROTOTYPE: $;$
1079 PPCODE:
1080{
1081 dREQ;
1082
1083 req->type = EIO_CLOSE;
1084 req->sv1 = newSVsv (fh);
1085 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087 EIO_SEND (req);
1088}
1089
1090void
1091aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092 ALIAS:
1093 aio_read = EIO_READ
1094 aio_write = EIO_WRITE
1095 PROTOTYPE: $$$$$;$
1096 PPCODE:
1097{
1098 STRLEN svlen;
1099 char *svptr = SvPVbyte (data, svlen);
1100 UV len = SvUV (length);
1101
1102 SvUPGRADE (data, SVt_PV);
1103 SvPOK_on (data);
1104
1105 if (dataoffset < 0)
1106 dataoffset += svlen;
1107
1108 if (dataoffset < 0 || dataoffset > svlen)
1109 croak ("dataoffset outside of data scalar");
1110
1111 if (ix == EIO_WRITE)
1112 {
1113 /* write: check length and adjust. */
1114 if (!SvOK (length) || len + dataoffset > svlen)
1115 len = svlen - dataoffset;
1116 }
1117 else
1118 {
1119 /* read: grow scalar as necessary */
1120 svptr = SvGROW (data, len + dataoffset + 1);
1121 }
1122
1123 if (len < 0)
1124 croak ("length must not be negative");
1125
1126 {
1127 dREQ;
1128
1129 req->type = ix;
1130 req->sv1 = newSVsv (fh);
1131 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132 : IoOFP (sv_2io (fh)));
1133 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134 req->size = len;
1135 req->sv2 = SvREFCNT_inc (data);
1136 req->ptr2 = (char *)svptr + dataoffset;
1137 req->stroffset = dataoffset;
1138
1139 if (!SvREADONLY (data))
1140 {
1141 SvREADONLY_on (data);
1142 req->flags |= FLAG_SV2_RO_OFF;
1143 }
1144
1145 EIO_SEND;
1146 }
1147}
1148
1149void
1150aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151 PROTOTYPE: $$;$
1152 PPCODE:
1153{
1154 SV *data;
1155 dREQ;
1156
1157 data = newSV (NAME_MAX);
1158 SvPOK_on (data);
1159
1160 req->type = EIO_READLINK;
1161 req->sv1 = newSVsv (path);
1162 req->ptr1 = SvPVbyte_nolen (req->sv1);
1163 req->sv2 = data;
1164 req->ptr2 = SvPVbyte_nolen (data);
1165
1166 EIO_SEND;
1167}
1168
1169void
1170aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171 PROTOTYPE: $$$$;$
1172 PPCODE:
1173{
1174 dREQ;
1175
1176 req->type = EIO_SENDFILE;
1177 req->sv1 = newSVsv (out_fh);
1178 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179 req->sv2 = newSVsv (in_fh);
1180 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181 req->offs = SvVAL64 (in_offset);
1182 req->size = length;
1183
1184 EIO_SEND;
1185}
1186
1187void
1188aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189 PROTOTYPE: $$$;$
1190 PPCODE:
1191{
1192 dREQ;
1193
1194 req->type = EIO_READAHEAD;
1195 req->sv1 = newSVsv (fh);
1196 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197 req->offs = SvVAL64 (offset);
1198 req->size = length;
1199
1200 EIO_SEND;
1201}
1202
1203void
1204aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205 ALIAS:
1206 aio_stat = EIO_STAT
1207 aio_lstat = EIO_LSTAT
1208 PPCODE:
1209{
1210 dREQ;
1211
1212 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213 if (!req->ptr2)
1214 {
1215 req_destroy (req);
1216 croak ("out of memory during aio_stat statdata allocation");
1217 }
1218
1219 req->flags |= FLAG_PTR2_FREE;
1220 req->sv1 = newSVsv (fh_or_path);
1221
1222 if (SvPOK (fh_or_path))
1223 {
1224 req->type = ix;
1225 req->ptr1 = SvPVbyte_nolen (req->sv1);
1226 }
1227 else
1228 {
1229 req->type = EIO_FSTAT;
1230 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231 }
1232
1233 EIO_SEND;
1234}
1235
1236void
1237aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238 PPCODE:
1239{
1240 dREQ;
1241
1242 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244 req->sv1 = newSVsv (fh_or_path);
1245
1246 if (SvPOK (fh_or_path))
1247 {
1248 req->type = EIO_UTIME;
1249 req->ptr1 = SvPVbyte_nolen (req->sv1);
1250 }
1251 else
1252 {
1253 req->type = EIO_FUTIME;
1254 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255 }
1256
1257 EIO_SEND;
1258}
1259
1260void
1261aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262 PPCODE:
1263{
1264 dREQ;
1265
1266 req->sv1 = newSVsv (fh_or_path);
1267 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269 if (SvPOK (fh_or_path))
1270 {
1271 req->type = EIO_TRUNCATE;
1272 req->ptr1 = SvPVbyte_nolen (req->sv1);
1273 }
1274 else
1275 {
1276 req->type = EIO_FTRUNCATE;
1277 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278 }
1279
1280 EIO_SEND;
1281}
1282
1283void
1284aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285 ALIAS:
1286 aio_chmod = EIO_CHMOD
1287 aio_fchmod = EIO_FCHMOD
1288 aio_mkdir = EIO_MKDIR
1289 PPCODE:
1290{
1291 dREQ;
1292
1293 req->type = type;
1294 req->int2 = mode;
1295 req->sv1 = newSVsv (fh_or_path);
1296
1297 if (ix == EIO_FCHMOD)
1298 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299 else
1300 req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302 EIO_SEND;
1303}
1304
1305void
1306aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307 PPCODE:
1308{
1309 dREQ;
1310
1311 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313 req->sv1 = newSVsv (fh_or_path);
1314
1315 if (SvPOK (fh_or_path))
1316 {
1317 req->type = EIO_CHOWN;
1318 req->ptr1 = SvPVbyte_nolen (req->sv1);
1319 }
1320 else
1321 {
1322 req->type = EIO_FCHOWN;
1323 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324 }
1325
1326 EIO_SEND;
1327}
1328
1329void
1330aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331 ALIAS:
1332 aio_unlink = EIO_UNLINK
1333 aio_rmdir = EIO_RMDIR
1334 aio_readdir = EIO_READDIR
1335 PPCODE:
1336{
1337 dREQ;
1338
1339 req->type = ix;
1340 req->sv1 = newSVsv (pathname);
1341 req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343 EIO_SEND;
1344}
1345
1346void
1347aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348 ALIAS:
1349 aio_link = EIO_LINK
1350 aio_symlink = EIO_SYMLINK
1351 aio_rename = EIO_RENAME
1352 PPCODE:
1353{
1354 dREQ;
1355
1356 req->type = ix;
1357 req->sv1 = newSVsv (oldpath);
1358 req->ptr1 = SvPVbyte_nolen (req->sv1);
1359 req->sv2 = newSVsv (newpath);
1360 req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362 EIO_SEND;
1363}
1364
1365void
1366aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367 PPCODE:
1368{
1369 dREQ;
1370
1371 req->type = EIO_MKNOD;
1372 req->sv1 = newSVsv (pathname);
1373 req->ptr1 = SvPVbyte_nolen (req->sv1);
1374 req->int2 = (mode_t)mode;
1375 req->offs = dev;
1376
1377 EIO_SEND;
1378}
1379
1380void
1381aio_busy (double delay, SV *callback=&PL_sv_undef)
1382 PPCODE:
1383{
1384 dREQ;
1385
1386 req->type = EIO_BUSY;
1387 req->nv1 = delay < 0. ? 0. : delay;
1388
1389 EIO_SEND;
1390}
1391
1392void
1393aio_group (SV *callback=&PL_sv_undef)
1394 PROTOTYPE: ;$
1395 PPCODE:
1396{
1397 dREQ;
1398
1399 req->type = EIO_GROUP;
1400
1401 req_send (req);
1402 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403}
1404
1405void
1406aio_nop (SV *callback=&PL_sv_undef)
1407 ALIAS:
1408 aio_nop = EIO_NOP
1409 aio_sync = EIO_SYNC
1410 PPCODE:
1411{
1412 dREQ;
1413
1414#endif
1415
1416void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2427eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417{ 2428{
1418 grp->int2 = limit; 2429 grp->int2 = limit;
1419 grp->feed = feed; 2430 grp->feed = feed;
1420 2431
1421 grp_try_feed (grp); 2432 grp_try_feed (grp);
1422} 2433}
1423 2434
2435void
1424void eio_grp_limit (eio_req *grp, int limit) 2436eio_grp_limit (eio_req *grp, int limit)
1425{ 2437{
1426 grp->int2 = limit; 2438 grp->int2 = limit;
1427 2439
1428 grp_try_feed (grp); 2440 grp_try_feed (grp);
1429} 2441}
1430 2442
2443void
1431void eio_grp_add (eio_req *grp, eio_req *req) 2444eio_grp_add (eio_req *grp, eio_req *req)
1432{ 2445{
1433 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2446 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2447
2448 grp->flags |= EIO_FLAG_GROUPADD;
1434 2449
1435 ++grp->size; 2450 ++grp->size;
1436 req->grp = grp; 2451 req->grp = grp;
1437 2452
1438 req->grp_prev = 0; 2453 req->grp_prev = 0;
1442 grp->grp_first->grp_prev = req; 2457 grp->grp_first->grp_prev = req;
1443 2458
1444 grp->grp_first = req; 2459 grp->grp_first = req;
1445} 2460}
1446 2461
2462/*****************************************************************************/
2463/* misc garbage */
1447 2464
2465eio_ssize_t
2466eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2467{
2468 return eio__sendfile (ofd, ifd, offset, count);
2469}
2470

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines