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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 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
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
18 81
19#ifndef EIO_FINISH 82#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 84#endif
22 85
… …
26 89
27#ifndef EIO_FEED 90#ifndef EIO_FEED
28# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
29#endif 92#endif
30 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
31#ifdef _WIN32 105#ifdef _WIN32
32 106
33 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #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 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu)
136
137 static int
138 eio__rename (const char *old, const char *neu)
139 {
140 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
141 return 0;
142
143 /* should steal _dosmaperr */
144 switch (GetLastError ())
145 {
146 case ERROR_FILE_NOT_FOUND:
147 case ERROR_PATH_NOT_FOUND:
148 case ERROR_INVALID_DRIVE:
149 case ERROR_NO_MORE_FILES:
150 case ERROR_BAD_NETPATH:
151 case ERROR_BAD_NET_NAME:
152 case ERROR_BAD_PATHNAME:
153 case ERROR_FILENAME_EXCED_RANGE:
154 errno = ENOENT;
155 break;
156
157 default:
158 errno = EACCES;
159 break;
160 }
161
162 return -1;
163 }
164
165 /* we could even stat and see if it exists */
166 static int
167 symlink (const char *old, const char *neu)
168 {
169 #if WINVER >= 0x0600
170 if (CreateSymbolicLink (neu, old, 1))
171 return 0;
172
173 if (CreateSymbolicLink (neu, old, 0))
174 return 0;
175 #endif
176
177 return EIO_ERRNO (ENOENT, -1);
178 }
179
180 /* POSIX API only */
181 #define CreateHardLink(neu,old,flags) 0
182 #define CreateSymbolicLink(neu,old,flags) 0
183
184 struct statvfs
185 {
186 int dummy;
187 };
188
189 #define DT_DIR EIO_DT_DIR
190 #define DT_REG EIO_DT_REG
191 #define D_NAME(entp) entp.cFileName
192 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 193
35#else 194#else
36 195
37# include "config.h"
38# include <sys/time.h> 196 #include <sys/time.h>
39# include <sys/select.h> 197 #include <sys/select.h>
198 #include <sys/statvfs.h>
40# include <unistd.h> 199 #include <unistd.h>
200 #include <signal.h>
201 #include <dirent.h>
202
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h>
205 #endif
206
207 #define D_NAME(entp) entp->d_name
208
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino
216 #endif
217
218 #ifdef _D_EXACT_NAMLEN
219 #undef D_NAMLEN
220 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
221 #endif
222
223 #ifdef _DIRENT_HAVE_D_TYPE
224 #define D_TYPE(de) (de)->d_type
225 #endif
226
227 #ifndef EIO_STRUCT_DIRENT
228 #define EIO_STRUCT_DIRENT struct dirent
229 #endif
230
231#endif
232
233#if HAVE_UTIMES
41# include <utime.h> 234# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 235#endif
47 236
48#endif 237#if HAVE_SYS_SYSCALL_H
49 238# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 239#endif
240
241#if HAVE_SYS_PRCTL_H
242# include <sys/prctl.h>
243#endif
53 244
54#if HAVE_SENDFILE 245#if HAVE_SENDFILE
55# if __linux 246# if __linux
56# include <sys/sendfile.h> 247# include <sys/sendfile.h>
57# elif __freebsd 248# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 249# include <sys/socket.h>
59# include <sys/uio.h> 250# include <sys/uio.h>
60# elif __hpux 251# elif __hpux
61# include <sys/socket.h> 252# include <sys/socket.h>
62# elif __solaris /* not yet */ 253# elif __solaris
63# include <sys/sendfile.h> 254# include <sys/sendfile.h>
64# else 255# else
65# error sendfile support requested but not available 256# error sendfile support requested but not available
66# endif 257# endif
67#endif 258#endif
68 259
69/* number of seconds after which an idle threads exit */ 260#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 261# define D_TYPE(de) 0
262#endif
263#ifndef D_INO
264# define D_INO(de) 0
265#endif
266#ifndef D_NAMLEN
267# define D_NAMLEN(entp) strlen (D_NAME (entp))
268#endif
71 269
72/* used for struct dirent, AIX doesn't provide it */ 270/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 271#ifndef NAME_MAX
74# define NAME_MAX 4096 272# define NAME_MAX 4096
75#endif 273#endif
76 274
275/* used for readlink etc. */
276#ifndef PATH_MAX
277# define PATH_MAX 4096
278#endif
279
77/* buffer size for various temporary buffers */ 280/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 281#define EIO_BUFSIZE 65536
79 282
80#define dBUF \ 283#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 284 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 285 errno = ENOMEM; \
85 if (!eio_buf) \ 286 if (!eio_buf) \
86 return -1; 287 return -1
288
289#define FUBd \
290 free (eio_buf)
87 291
88#define EIO_TICKS ((1000000 + 1023) >> 10) 292#define EIO_TICKS ((1000000 + 1023) >> 10)
293
294#define ETP_PRI_MIN EIO_PRI_MIN
295#define ETP_PRI_MAX EIO_PRI_MAX
296
297struct etp_worker;
298
299#define ETP_REQ eio_req
300#define ETP_DESTROY(req) eio_destroy (req)
301static int eio_finish (eio_req *req);
302#define ETP_FINISH(req) eio_finish (req)
303static void eio_execute (struct etp_worker *self, eio_req *req);
304#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
305
306/*****************************************************************************/
307
308#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
309
310/* calculate time difference in ~1/EIO_TICKS of a second */
311ecb_inline int
312tvdiff (struct timeval *tv1, struct timeval *tv2)
313{
314 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
315 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
316}
317
318static unsigned int started, idle, wanted = 4;
89 319
90static void (*want_poll_cb) (void); 320static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 321static void (*done_poll_cb) (void);
92 322
93static unsigned int max_poll_time = 0; 323static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 324static unsigned int max_poll_reqs; /* reslock */
95 325
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 326static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 327static volatile unsigned int nready; /* reqlock */
98{ 328static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 329static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 330static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 331
103static unsigned int started, idle, wanted; 332static xmutex_t wrklock;
333static xmutex_t reslock;
334static xmutex_t reqlock;
335static xcond_t reqwait;
336
337#if !HAVE_PREADWRITE
338/*
339 * make our pread/pwrite emulation safe against themselves, but not against
340 * normal read/write by using a mutex. slows down execution a lot,
341 * but that's your problem, not mine.
342 */
343static xmutex_t preadwritelock;
344#endif
345
346typedef struct etp_worker
347{
348 /* locked by wrklock */
349 struct etp_worker *prev, *next;
350
351 xthread_t tid;
352
353 /* locked by reslock, reqlock or wrklock */
354 ETP_REQ *req; /* currently processed request */
355
356#ifdef ETP_WORKER_COMMON
357 ETP_WORKER_COMMON
358#endif
359} etp_worker;
360
361static etp_worker wrk_first; /* NOT etp */
362
363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 365
105/* worker threads management */ 366/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 367
108typedef struct worker { 368static void ecb_cold
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 369etp_worker_clear (etp_worker *wrk)
123{ 370{
124 if (wrk->dirp)
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 371}
136 372
137static void worker_free (worker *wrk) 373static void ecb_cold
374etp_worker_free (etp_worker *wrk)
138{ 375{
139 wrk->next->prev = wrk->prev; 376 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 377 wrk->prev->next = wrk->next;
141 378
142 free (wrk); 379 free (wrk);
143} 380}
144 381
145static volatile unsigned int nreqs, nready, npending; 382static unsigned int
146static volatile unsigned int max_idle = 4; 383etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 384{
385 int retval;
386 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
387 retval = nreqs;
388 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 389 return retval;
155} 390}
156 391
157unsigned int eio_nready (void) 392static unsigned int
393etp_nready (void)
158{ 394{
159 unsigned int retval; 395 unsigned int retval;
160 396
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 397 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 398 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 399 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 400
165 return retval; 401 return retval;
166} 402}
167 403
168unsigned int eio_npending (void) 404static unsigned int
405etp_npending (void)
169{ 406{
170 unsigned int retval; 407 unsigned int retval;
171 408
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 409 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 410 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 411 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 412
176 return retval; 413 return retval;
177} 414}
178 415
179unsigned int eio_nthreads (void) 416static unsigned int
417etp_nthreads (void)
180{ 418{
181 unsigned int retval; 419 unsigned int retval;
182 420
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 421 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 422 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 429 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 430 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 431 * per shift, the most expensive operation.
194 */ 432 */
195typedef struct { 433typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 434 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 435 int size;
198} reqq; 436} etp_reqq;
199 437
200static reqq req_queue; 438static etp_reqq req_queue;
201static reqq res_queue; 439static etp_reqq res_queue;
202 440
203static int reqq_push (reqq *q, eio_req *req) 441static void ecb_noinline ecb_cold
442reqq_init (etp_reqq *q)
443{
444 int pri;
445
446 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
447 q->qs[pri] = q->qe[pri] = 0;
448
449 q->size = 0;
450}
451
452static int ecb_noinline
453reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 454{
205 int pri = req->pri; 455 int pri = req->pri;
206 req->next = 0; 456 req->next = 0;
207 457
208 if (q->qe[pri]) 458 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 464 q->qe[pri] = q->qs[pri] = req;
215 465
216 return q->size++; 466 return q->size++;
217} 467}
218 468
219static eio_req *reqq_shift (reqq *q) 469static ETP_REQ * ecb_noinline
470reqq_shift (etp_reqq *q)
220{ 471{
221 int pri; 472 int pri;
222 473
223 if (!q->size) 474 if (!q->size)
224 return 0; 475 return 0;
225 476
226 --q->size; 477 --q->size;
227 478
228 for (pri = EIO_NUM_PRI; pri--; ) 479 for (pri = ETP_NUM_PRI; pri--; )
229 { 480 {
230 eio_req *req = q->qs[pri]; 481 eio_req *req = q->qs[pri];
231 482
232 if (req) 483 if (req)
233 { 484 {
… …
239 } 490 }
240 491
241 abort (); 492 abort ();
242} 493}
243 494
244static void grp_try_feed (eio_req *grp) 495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 497{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 498 X_MUTEX_CREATE (wrklock);
247 { 499 X_MUTEX_CREATE (reslock);
248 int old_len = grp->size; 500 X_MUTEX_CREATE (reqlock);
501 X_COND_CREATE (reqwait);
249 502
250 EIO_FEED (grp); 503 reqq_init (&req_queue);
504 reqq_init (&res_queue);
251 505
252 /* stop if no progress has been made */ 506 wrk_first.next =
253 if (old_len == grp->size) 507 wrk_first.prev = &wrk_first;
254 {
255 grp->feed = 0;
256 break;
257 }
258 }
259}
260 508
261static int eio_finish (eio_req *req); 509 started = 0;
510 idle = 0;
511 nreqs = 0;
512 nready = 0;
513 npending = 0;
262 514
263static int grp_dec (eio_req *grp) 515 want_poll_cb = want_poll;
264{ 516 done_poll_cb = done_poll;
265 --grp->size;
266 517
267 /* call feeder, if applicable */
268 grp_try_feed (grp);
269
270 /* finish, if done */
271 if (!grp->size && grp->int1)
272 return eio_finish (grp);
273 else
274 return 0; 518 return 0;
275} 519}
276 520
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 521X_THREAD_PROC (etp_proc);
326 522
327static void start_thread (void) 523static void ecb_cold
524etp_start_thread (void)
328{ 525{
329 worker *wrk = calloc (1, sizeof (worker)); 526 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 527
331 if (!wrk) 528 /*TODO*/
332 croak ("unable to allocate worker thread data"); 529 assert (("unable to allocate worker thread data", wrk));
333 530
334 X_LOCK (wrklock); 531 X_LOCK (wrklock);
335 532
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 533 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 534 {
338 wrk->prev = &wrk_first; 535 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 536 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 537 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 538 wrk_first.next = wrk;
… …
345 free (wrk); 542 free (wrk);
346 543
347 X_UNLOCK (wrklock); 544 X_UNLOCK (wrklock);
348} 545}
349 546
547static void
350static void maybe_start_thread (void) 548etp_maybe_start_thread (void)
351{ 549{
352 if (eio_nthreads () >= wanted) 550 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 551 return;
354 552
355 /* todo: maybe use idle here, but might be less exact */ 553 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 554 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 555 return;
358 556
359 start_thread (); 557 etp_start_thread ();
360} 558}
361 559
362void eio_submit (eio_req *req) 560static void ecb_cold
363{ 561etp_end_thread (void)
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{ 562{
377 eio_req *req = calloc (1, sizeof (eio_req)); 563 eio_req *req = calloc (1, sizeof (eio_req));
378 564
379 req->type = EIO_QUIT; 565 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 566 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 567
382 X_LOCK (reqlock); 568 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 569 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 570 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 571 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 573 X_LOCK (wrklock);
388 --started; 574 --started;
389 X_UNLOCK (wrklock); 575 X_UNLOCK (wrklock);
390} 576}
391 577
392void eio_set_max_poll_time (double nseconds) 578static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 579etp_poll (void)
429{ 580{
430 int maxreqs = max_poll_reqs; 581 unsigned int maxreqs;
582 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 583 struct timeval tv_start, tv_now;
432 eio_req *req;
433 584
585 X_LOCK (reslock);
586 maxreqs = max_poll_reqs;
587 maxtime = max_poll_time;
588 X_UNLOCK (reslock);
589
434 if (max_poll_time) 590 if (maxtime)
435 gettimeofday (&tv_start, 0); 591 gettimeofday (&tv_start, 0);
436 592
437 for (;;) 593 for (;;)
438 { 594 {
595 ETP_REQ *req;
596
439 maybe_start_thread (); 597 etp_maybe_start_thread ();
440 598
441 X_LOCK (reslock); 599 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 600 req = reqq_shift (&res_queue);
443 601
444 if (req) 602 if (req)
… …
452 X_UNLOCK (reslock); 610 X_UNLOCK (reslock);
453 611
454 if (!req) 612 if (!req)
455 return 0; 613 return 0;
456 614
615 X_LOCK (reqlock);
457 --nreqs; 616 --nreqs;
617 X_UNLOCK (reqlock);
458 618
459 if (req->type == EIO_GROUP && req->size) 619 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 620 {
461 req->int1 = 1; /* mark request as delayed */ 621 req->int1 = 1; /* mark request as delayed */
462 continue; 622 continue;
463 } 623 }
464 else 624 else
465 { 625 {
466 int res = eio_finish (req); 626 int res = ETP_FINISH (req);
467 if (res) 627 if (ecb_expect_false (res))
468 return res; 628 return res;
469 } 629 }
470 630
471 if (maxreqs && !--maxreqs) 631 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 632 break;
473 633
474 if (max_poll_time) 634 if (maxtime)
475 { 635 {
476 gettimeofday (&tv_now, 0); 636 gettimeofday (&tv_now, 0);
477 637
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 638 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 639 break;
480 } 640 }
481 } 641 }
482 642
483 errno = EAGAIN; 643 errno = EAGAIN;
484 return -1; 644 return -1;
485} 645}
486 646
647static void
648etp_cancel (ETP_REQ *req)
649{
650 req->cancelled = 1;
651
652 eio_grp_cancel (req);
653}
654
655static void
656etp_submit (ETP_REQ *req)
657{
658 req->pri -= ETP_PRI_MIN;
659
660 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
661 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
662
663 if (ecb_expect_false (req->type == EIO_GROUP))
664 {
665 /* I hope this is worth it :/ */
666 X_LOCK (reqlock);
667 ++nreqs;
668 X_UNLOCK (reqlock);
669
670 X_LOCK (reslock);
671
672 ++npending;
673
674 if (!reqq_push (&res_queue, req) && want_poll_cb)
675 want_poll_cb ();
676
677 X_UNLOCK (reslock);
678 }
679 else
680 {
681 X_LOCK (reqlock);
682 ++nreqs;
683 ++nready;
684 reqq_push (&req_queue, req);
685 X_COND_SIGNAL (reqwait);
686 X_UNLOCK (reqlock);
687
688 etp_maybe_start_thread ();
689 }
690}
691
692static void ecb_cold
693etp_set_max_poll_time (double nseconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
696 max_poll_time = nseconds * EIO_TICKS;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
698}
699
700static void ecb_cold
701etp_set_max_poll_reqs (unsigned int maxreqs)
702{
703 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
704 max_poll_reqs = maxreqs;
705 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
706}
707
708static void ecb_cold
709etp_set_max_idle (unsigned int nthreads)
710{
711 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
712 max_idle = nthreads;
713 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
714}
715
716static void ecb_cold
717etp_set_idle_timeout (unsigned int seconds)
718{
719 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
720 idle_timeout = seconds;
721 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
722}
723
724static void ecb_cold
725etp_set_min_parallel (unsigned int nthreads)
726{
727 if (wanted < nthreads)
728 wanted = nthreads;
729}
730
731static void ecb_cold
732etp_set_max_parallel (unsigned int nthreads)
733{
734 if (wanted > nthreads)
735 wanted = nthreads;
736
737 while (started > wanted)
738 etp_end_thread ();
739}
740
741/*****************************************************************************/
742
743static void
744grp_try_feed (eio_req *grp)
745{
746 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
747 {
748 grp->flags &= ~EIO_FLAG_GROUPADD;
749
750 EIO_FEED (grp);
751
752 /* stop if no progress has been made */
753 if (!(grp->flags & EIO_FLAG_GROUPADD))
754 {
755 grp->feed = 0;
756 break;
757 }
758 }
759}
760
761static int
762grp_dec (eio_req *grp)
763{
764 --grp->size;
765
766 /* call feeder, if applicable */
767 grp_try_feed (grp);
768
769 /* finish, if done */
770 if (!grp->size && grp->int1)
771 return eio_finish (grp);
772 else
773 return 0;
774}
775
776static void
777eio_destroy (eio_req *req)
778{
779 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
780 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
781
782 EIO_DESTROY (req);
783}
784
785static int
786eio_finish (eio_req *req)
787{
788 int res = EIO_FINISH (req);
789
790 if (req->grp)
791 {
792 int res2;
793 eio_req *grp = req->grp;
794
795 /* unlink request */
796 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
797 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
798
799 if (grp->grp_first == req)
800 grp->grp_first = req->grp_next;
801
802 res2 = grp_dec (grp);
803
804 if (!res)
805 res = res2;
806 }
807
808 eio_destroy (req);
809
810 return res;
811}
812
813void
814eio_grp_cancel (eio_req *grp)
815{
816 for (grp = grp->grp_first; grp; grp = grp->grp_next)
817 eio_cancel (grp);
818}
819
820void
821eio_cancel (eio_req *req)
822{
823 etp_cancel (req);
824}
825
826void
827eio_submit (eio_req *req)
828{
829 etp_submit (req);
830}
831
832unsigned int
833eio_nreqs (void)
834{
835 return etp_nreqs ();
836}
837
838unsigned int
839eio_nready (void)
840{
841 return etp_nready ();
842}
843
844unsigned int
845eio_npending (void)
846{
847 return etp_npending ();
848}
849
850unsigned int ecb_cold
851eio_nthreads (void)
852{
853 return etp_nthreads ();
854}
855
856void ecb_cold
857eio_set_max_poll_time (double nseconds)
858{
859 etp_set_max_poll_time (nseconds);
860}
861
862void ecb_cold
863eio_set_max_poll_reqs (unsigned int maxreqs)
864{
865 etp_set_max_poll_reqs (maxreqs);
866}
867
868void ecb_cold
869eio_set_max_idle (unsigned int nthreads)
870{
871 etp_set_max_idle (nthreads);
872}
873
874void ecb_cold
875eio_set_idle_timeout (unsigned int seconds)
876{
877 etp_set_idle_timeout (seconds);
878}
879
880void ecb_cold
881eio_set_min_parallel (unsigned int nthreads)
882{
883 etp_set_min_parallel (nthreads);
884}
885
886void ecb_cold
887eio_set_max_parallel (unsigned int nthreads)
888{
889 etp_set_max_parallel (nthreads);
890}
891
892int eio_poll (void)
893{
894 return etp_poll ();
895}
896
487/*****************************************************************************/ 897/*****************************************************************************/
488/* work around various missing functions */ 898/* work around various missing functions */
489 899
490#if !HAVE_PREADWRITE 900#if !HAVE_PREADWRITE
901# undef pread
902# undef pwrite
491# define pread aio_pread 903# define pread eio__pread
492# define pwrite aio_pwrite 904# define pwrite eio__pwrite
493 905
494/* 906static eio_ssize_t
495 * make our pread/pwrite safe against themselves, but not against 907eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 908{
503 ssize_t res; 909 eio_ssize_t res;
504 off_t ooffset; 910 off_t ooffset;
505 911
506 X_LOCK (preadwritelock); 912 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR); 913 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET); 914 lseek (fd, offset, SEEK_SET);
… …
511 X_UNLOCK (preadwritelock); 917 X_UNLOCK (preadwritelock);
512 918
513 return res; 919 return res;
514} 920}
515 921
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 922static eio_ssize_t
923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 924{
518 ssize_t res; 925 eio_ssize_t res;
519 off_t ooffset; 926 off_t ooffset;
520 927
521 X_LOCK (preadwritelock); 928 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 929 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 930 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 931 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 932 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 933 X_UNLOCK (preadwritelock);
527 934
528 return res; 935 return res;
529} 936}
530#endif 937#endif
531 938
532#ifndef HAVE_FUTIMES 939#ifndef HAVE_UTIMES
533 940
941# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 942# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 943
944static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 945eio__utimes (const char *filename, const struct timeval times[2])
538{ 946{
539 if (times) 947 if (times)
540 { 948 {
541 struct utimbuf buf; 949 struct utimbuf buf;
542 950
… …
547 } 955 }
548 else 956 else
549 return utime (filename, 0); 957 return utime (filename, 0);
550} 958}
551 959
960#endif
961
962#ifndef HAVE_FUTIMES
963
964# undef futimes
965# define futimes(fd,times) eio__futimes (fd, times)
966
967static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 968eio__futimes (int fd, const struct timeval tv[2])
553{ 969{
554 errno = ENOSYS; 970 errno = ENOSYS;
555 return -1; 971 return -1;
556} 972}
557 973
558#endif 974#endif
559 975
560#if !HAVE_FDATASYNC 976#if !HAVE_FDATASYNC
977# undef fdatasync
561# define fdatasync fsync 978# define fdatasync(fd) fsync (fd)
562#endif 979#endif
980
981static int
982eio__syncfs (int fd)
983{
984 int res;
985
986#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else
989 res = -1;
990 errno = ENOSYS;
991#endif
992
993 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync ();
995
996 return res;
997}
998
999/* sync_file_range always needs emulation */
1000static int
1001eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1002{
1003#if HAVE_SYNC_FILE_RANGE
1004 int res;
1005
1006 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1007 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1008 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1009 {
1010 flags = 0
1011 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1012 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1013 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1014 }
1015
1016 res = sync_file_range (fd, offset, nbytes, flags);
1017
1018 if (!res || errno != ENOSYS)
1019 return res;
1020#endif
1021
1022 /* even though we could play tricks with the flags, it's better to always
1023 * call fdatasync, as that matches the expectation of its users best */
1024 return fdatasync (fd);
1025}
1026
1027static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{
1030#if HAVE_FALLOCATE
1031 return fallocate (fd, mode, offset, len);
1032#else
1033 errno = ENOSYS;
1034 return -1;
1035#endif
1036}
563 1037
564#if !HAVE_READAHEAD 1038#if !HAVE_READAHEAD
1039# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1040# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 1041
1042static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1043eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 1044{
569 size_t todo = count; 1045 size_t todo = count;
570 dBUF; 1046 dBUF;
571 1047
572 while (todo > 0) 1048 while (todo > 0)
573 { 1049 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1050 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 1051
576 pread (fd, aio_buf, len, offset); 1052 pread (fd, eio_buf, len, offset);
577 offset += len; 1053 offset += len;
578 todo -= len; 1054 todo -= len;
579 } 1055 }
580 1056
1057 FUBd;
1058
581 errno = 0; 1059 errno = 0;
582 return count; 1060 return count;
583} 1061}
584 1062
585#endif 1063#endif
586 1064
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 1065/* sendfile always needs emulation */
1066static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1067eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 1068{
1069 eio_ssize_t written = 0;
620 ssize_t res; 1070 eio_ssize_t res;
621 1071
622 if (!count) 1072 if (!count)
623 return 0; 1073 return 0;
624 1074
1075 for (;;)
1076 {
1077#ifdef __APPLE__
1078# undef HAVE_SENDFILE /* broken, as everything on os x */
1079#endif
625#if HAVE_SENDFILE 1080#if HAVE_SENDFILE
626# if __linux 1081# if __linux
1082 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1083 res = sendfile (ofd, ifd, &soffset, count);
628 1084
629# elif __freebsd 1085# elif __FreeBSD__
630 /* 1086 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1087 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1088 * wasted on making it similar to other I/O functions.
633 */ 1089 */
634 {
635 off_t sbytes; 1090 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1091 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1092
638 if (res < 0 && sbytes) 1093 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1094 /* freebsd' sendfile will return 0 on success */
1095 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1096 /* not on e.g. EIO or EPIPE - sounds broken */
1097 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1098 res = sbytes;
641 } 1099 #endif
1100
1101 /* according to source inspection, this is correct, and useful behaviour */
1102 if (sbytes)
1103 res = sbytes;
1104
1105# elif defined (__APPLE__)
1106 off_t sbytes = count;
1107 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1108
1109 /* according to the manpage, sbytes is always valid */
1110 if (sbytes)
1111 res = sbytes;
642 1112
643# elif __hpux 1113# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1114 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1115
646# elif __solaris 1116# elif __solaris
647 {
648 struct sendfilevec vec; 1117 struct sendfilevec vec;
649 size_t sbytes; 1118 size_t sbytes;
650 1119
651 vec.sfv_fd = ifd; 1120 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1121 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1122 vec.sfv_off = offset;
654 vec.sfv_len = count; 1123 vec.sfv_len = count;
655 1124
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1125 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1126
658 if (res < 0 && sbytes) 1127 if (res < 0 && sbytes)
659 res = sbytes; 1128 res = sbytes;
660 }
661 1129
662# endif 1130# endif
1131
1132#elif defined (_WIN32) && 0
1133 /* does not work, just for documentation of what would need to be done */
1134 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1135 /* libeio guarantees that the file offset does not change, and windows */
1136 /* has no way to get an independent handle to the same file description */
1137 HANDLE h = TO_SOCKET (ifd);
1138 SetFilePointer (h, offset, 0, FILE_BEGIN);
1139 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1140
663#else 1141#else
664 res = -1; 1142 res = -1;
665 errno = ENOSYS; 1143 errno = ENOSYS;
666#endif 1144#endif
667 1145
1146 /* we assume sendfile can copy at least 128mb in one go */
1147 if (res <= 128 * 1024 * 1024)
1148 {
1149 if (res > 0)
1150 written += res;
1151
1152 if (written)
1153 return written;
1154
1155 break;
1156 }
1157 else
1158 {
1159 /* if we requested more, then probably the kernel was lazy */
1160 written += res;
1161 offset += res;
1162 count -= res;
1163
1164 if (!count)
1165 return written;
1166 }
1167 }
1168
668 if (res < 0 1169 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1170 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1171 /* BSDs */
1172#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1173 || errno == ENOTSUP
1174#endif
1175#ifdef EOPNOTSUPP /* windows */
1176 || errno == EOPNOTSUPP /* BSDs */
1177#endif
670#if __solaris 1178#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1179 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1180#endif
673 ) 1181 )
674 ) 1182 )
… …
678 1186
679 res = 0; 1187 res = 0;
680 1188
681 while (count) 1189 while (count)
682 { 1190 {
683 ssize_t cnt; 1191 eio_ssize_t cnt;
684 1192
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1193 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 1194
687 if (cnt <= 0) 1195 if (cnt <= 0)
688 { 1196 {
… …
700 1208
701 offset += cnt; 1209 offset += cnt;
702 res += cnt; 1210 res += cnt;
703 count -= cnt; 1211 count -= cnt;
704 } 1212 }
1213
1214 FUBd;
705 } 1215 }
706 1216
707 return res; 1217 return res;
708} 1218}
709 1219
1220#ifdef PAGESIZE
1221# define eio_pagesize() PAGESIZE
1222#else
1223static intptr_t
1224eio_pagesize (void)
1225{
1226 static intptr_t page;
1227
1228 if (!page)
1229 page = sysconf (_SC_PAGESIZE);
1230
1231 return page;
1232}
1233#endif
1234
1235static void
1236eio_page_align (void **addr, size_t *length)
1237{
1238 intptr_t mask = eio_pagesize () - 1;
1239
1240 /* round down addr */
1241 intptr_t adj = mask & (intptr_t)*addr;
1242
1243 *addr = (void *)((intptr_t)*addr - adj);
1244 *length += adj;
1245
1246 /* round up length */
1247 *length = (*length + mask) & ~mask;
1248}
1249
1250#if !_POSIX_MEMLOCK
1251# define eio__mlockall(a) EIO_ENOSYS ()
1252#else
1253
1254static int
1255eio__mlockall (int flags)
1256{
1257 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1258 extern int mallopt (int, int);
1259 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1260 #endif
1261
1262 if (EIO_MCL_CURRENT != MCL_CURRENT
1263 || EIO_MCL_FUTURE != MCL_FUTURE)
1264 {
1265 flags = 0
1266 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1267 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1268 }
1269
1270 return mlockall (flags);
1271}
1272#endif
1273
1274#if !_POSIX_MEMLOCK_RANGE
1275# define eio__mlock(a,b) EIO_ENOSYS ()
1276#else
1277
1278static int
1279eio__mlock (void *addr, size_t length)
1280{
1281 eio_page_align (&addr, &length);
1282
1283 return mlock (addr, length);
1284}
1285
1286#endif
1287
1288#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1289# define eio__msync(a,b,c) EIO_ENOSYS ()
1290#else
1291
1292static int
1293eio__msync (void *mem, size_t len, int flags)
1294{
1295 eio_page_align (&mem, &len);
1296
1297 if (EIO_MS_ASYNC != MS_SYNC
1298 || EIO_MS_INVALIDATE != MS_INVALIDATE
1299 || EIO_MS_SYNC != MS_SYNC)
1300 {
1301 flags = 0
1302 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1303 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1304 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1305 }
1306
1307 return msync (mem, len, flags);
1308}
1309
1310#endif
1311
1312static int
1313eio__mtouch (eio_req *req)
1314{
1315 void *mem = req->ptr2;
1316 size_t len = req->size;
1317 int flags = req->int1;
1318
1319 eio_page_align (&mem, &len);
1320
1321 {
1322 intptr_t addr = (intptr_t)mem;
1323 intptr_t end = addr + len;
1324 intptr_t page = eio_pagesize ();
1325
1326 if (addr < end)
1327 if (flags & EIO_MT_MODIFY) /* modify */
1328 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1329 else
1330 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1331 }
1332
1333 return 0;
1334}
1335
1336/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */
1338
1339static void
1340eio__realpath (eio_req *req, etp_worker *self)
1341{
1342 char *rel = req->ptr1;
1343 char *res;
1344 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX;
1347#else
1348 int symlinks = 32;
1349#endif
1350
1351 req->result = -1;
1352
1353 errno = EINVAL;
1354 if (!rel)
1355 return;
1356
1357 errno = ENOENT;
1358 if (!*rel)
1359 return;
1360
1361 if (!req->ptr2)
1362 {
1363 X_LOCK (wrklock);
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return;
1371 }
1372
1373 res = req->ptr2;
1374 tmp1 = res + PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX;
1376
1377#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1379 /* on linux we may be able to ask the kernel */
1380 {
1381 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1382
1383 if (fd >= 0)
1384 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX);
1387 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */
1390
1391 if (req->result > 0)
1392 goto done;
1393 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return;
1396 }
1397#endif
1398#endif
1399
1400 if (*rel != '/')
1401 {
1402 if (!getcwd (res, PATH_MAX))
1403 return;
1404
1405 if (res [1]) /* only use if not / */
1406 res += strlen (res);
1407 }
1408
1409 while (*rel)
1410 {
1411 eio_ssize_t len, linklen;
1412 char *beg = rel;
1413
1414 while (*rel && *rel != '/')
1415 ++rel;
1416
1417 len = rel - beg;
1418
1419 if (!len) /* skip slashes */
1420 {
1421 ++rel;
1422 continue;
1423 }
1424
1425 if (beg [0] == '.')
1426 {
1427 if (len == 1)
1428 continue; /* . - nop */
1429
1430 if (beg [1] == '.' && len == 2)
1431 {
1432 /* .. - back up one component, if possible */
1433
1434 while (res != req->ptr2)
1435 if (*--res == '/')
1436 break;
1437
1438 continue;
1439 }
1440 }
1441
1442 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1)
1444 return;
1445
1446 /* copy one component */
1447 *res = '/';
1448 memcpy (res + 1, beg, len);
1449
1450 /* zero-terminate, for readlink */
1451 res [len + 1] = 0;
1452
1453 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1455
1456 if (linklen < 0)
1457 {
1458 if (errno != EINVAL)
1459 return;
1460
1461 /* it's a normal directory. hopefully */
1462 res += len + 1;
1463 }
1464 else
1465 {
1466 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel);
1468
1469 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX)
1471 return;
1472
1473 errno = ELOOP;
1474 if (!--symlinks)
1475 return;
1476
1477 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */
1479
1480 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen);
1484
1485 rel = tmp2;
1486 }
1487 }
1488
1489 /* special case for the lone root path */
1490 if (res == req->ptr2)
1491 *res++ = '/';
1492
1493 req->result = res - (char *)req->ptr2;
1494
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1497}
1498
1499static signed char
1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1501{
1502 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1503 : a->inode < b->inode ? -1
1504 : a->inode > b->inode ? 1
1505 : 0;
1506}
1507
1508#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1509
1510#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1511#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1512
1513static void
1514eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1515{
1516 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1517 unsigned char *bit = bits;
1518
1519 assert (CHAR_BIT == 8);
1520 assert (sizeof (eio_dirent) * 8 < 256);
1521 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1522 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1523
1524 if (size <= EIO_SORT_FAST)
1525 return;
1526
1527 /* first prepare an array of bits to test in our radix sort */
1528 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1529 /* inode_bits must contain all inodes ORed together */
1530 /* which is used to skip bits that are 0 everywhere, which is very common */
1531 {
1532 eio_ino_t endianness;
1533 int i, j;
1534
1535 /* we store the byte offset of byte n into byte n of "endianness" */
1536 for (i = 0; i < sizeof (eio_ino_t); ++i)
1537 ((unsigned char *)&endianness)[i] = i;
1538
1539 *bit++ = 0;
1540
1541 for (i = 0; i < sizeof (eio_ino_t); ++i)
1542 {
1543 /* shifting off the byte offsets out of "endianness" */
1544 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1545 endianness >>= 8;
1546
1547 for (j = 0; j < 8; ++j)
1548 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1549 *bit++ = offs + j;
1550 }
1551
1552 for (j = 0; j < 8; ++j)
1553 if (score_bits & (1 << j))
1554 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1555 }
1556
1557 /* now actually do the sorting (a variant of MSD radix sort) */
1558 {
1559 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1560 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1561 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1562 int stk_idx = 0;
1563
1564 base_stk [stk_idx] = dents;
1565 end_stk [stk_idx] = dents + size;
1566 bit_stk [stk_idx] = bit - 1;
1567
1568 do
1569 {
1570 base = base_stk [stk_idx];
1571 end = end_stk [stk_idx];
1572 bit = bit_stk [stk_idx];
1573
1574 for (;;)
1575 {
1576 unsigned char O = *bit >> 3;
1577 unsigned char M = 1 << (*bit & 7);
1578
1579 eio_dirent *a = base;
1580 eio_dirent *b = end;
1581
1582 if (b - a < EIO_SORT_CUTOFF)
1583 break;
1584
1585 /* now bit-partition the array on the bit */
1586 /* this ugly asymmetric loop seems to perform much better than typical */
1587 /* partition algos found in the literature */
1588 do
1589 if (!(((unsigned char *)a)[O] & M))
1590 ++a;
1591 else if (!(((unsigned char *)--b)[O] & M))
1592 {
1593 eio_dirent tmp = *a; *a = *b; *b = tmp;
1594 ++a;
1595 }
1596 while (b > a);
1597
1598 /* next bit, or stop, if no bits left in this path */
1599 if (!*--bit)
1600 break;
1601
1602 base_stk [stk_idx] = a;
1603 end_stk [stk_idx] = end;
1604 bit_stk [stk_idx] = bit;
1605 ++stk_idx;
1606
1607 end = a;
1608 }
1609 }
1610 while (stk_idx--);
1611 }
1612}
1613
1614static void
1615eio_dent_insertion_sort (eio_dirent *dents, int size)
1616{
1617 /* first move the smallest element to the front, to act as a sentinel */
1618 {
1619 int i;
1620 eio_dirent *min = dents;
1621
1622 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1623 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1624 if (EIO_DENT_CMP (dents [i], <, *min))
1625 min = &dents [i];
1626
1627 /* swap elements 0 and j (minimum) */
1628 {
1629 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1630 }
1631 }
1632
1633 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1634 {
1635 eio_dirent *i, *j;
1636
1637 for (i = dents + 1; i < dents + size; ++i)
1638 {
1639 eio_dirent value = *i;
1640
1641 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1642 j [1] = j [0];
1643
1644 j [1] = value;
1645 }
1646 }
1647}
1648
1649static void
1650eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1651{
1652 if (size <= 1)
1653 return; /* our insertion sort relies on size > 0 */
1654
1655 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1656 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1657 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1658
1659 /* use an insertion sort at the end, or for small arrays, */
1660 /* as insertion sort is more efficient for small partitions */
1661 eio_dent_insertion_sort (dents, size);
1662}
1663
710/* read a full directory */ 1664/* read a full directory */
1665static void
711static void scandir_ (eio_req *req, worker *self) 1666eio__scandir (eio_req *req, etp_worker *self)
712{ 1667{
1668 char *name, *names;
1669 int namesalloc = 4096 - sizeof (void *) * 4;
1670 int namesoffs = 0;
1671 int flags = req->int1;
1672 eio_dirent *dents = 0;
1673 int dentalloc = 128;
1674 int dentoffs = 0;
1675 eio_ino_t inode_bits = 0;
1676#ifdef _WIN32
1677 HANDLE dirp;
1678 WIN32_FIND_DATA entp;
1679#else
713 DIR *dirp; 1680 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1681 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1682#endif
721 int memlen = 4096;
722 int memofs = 0;
723 int res = 0;
724 1683
725 X_LOCK (wrklock); 1684 req->result = -1;
1685
1686 if (!(flags & EIO_READDIR_DENTS))
1687 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1688
1689#ifdef _WIN32
1690 {
1691 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH);
1693 const char *fmt;
1694
1695 if (!len)
1696 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1698 fmt = "%s*";
1699 else
1700 fmt = "%s/*";
1701
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1703 dirp = FindFirstFile (path, &entp);
1704 free (path);
1705
1706 if (dirp == INVALID_HANDLE_VALUE)
1707 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */
1711 switch (GetLastError ())
1712 {
1713 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0;
1715 break;
1716
1717 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT;
1721 break;
1722
1723 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM;
1725 break;
1726
1727 default:
1728 errno = EINVAL;
1729 break;
1730 }
1731 }
1732 }
1733#else
726 self->dirp = dirp = opendir (req->ptr1); 1734 dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1735#endif
1736
728 req->flags |= EIO_FLAG_PTR2_FREE; 1737 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1);
1739
1740 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1742 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock);
731 1743
732 if (dirp && u && names) 1744 if (dirp && names && (!flags || dents))
733 for (;;) 1745 for (;;)
734 { 1746 {
1747 int done;
1748
1749#ifdef _WIN32
1750 done = !dirp;
1751#else
735 errno = 0; 1752 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1753 entp = readdir (dirp);
1754 done = !entp;
1755#endif
737 1756
738 if (!entp) 1757 if (done)
1758 {
1759#ifndef _WIN32
1760 int old_errno = errno;
1761 closedir (dirp);
1762 errno = old_errno;
1763
1764 if (errno)
1765 break;
1766#endif
1767
1768 /* sort etc. */
1769 req->int1 = flags;
1770 req->result = dentoffs;
1771
1772 if (flags & EIO_READDIR_STAT_ORDER)
1773 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1774 else if (flags & EIO_READDIR_DIRS_FIRST)
1775 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1776 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1777 else
1778 {
1779 /* in this case, all is known, and we just put dirs first and sort them */
1780 eio_dirent *oth = dents + dentoffs;
1781 eio_dirent *dir = dents;
1782
1783 /* now partition dirs to the front, and non-dirs to the back */
1784 /* by walking from both sides and swapping if necessary */
1785 while (oth > dir)
1786 {
1787 if (dir->type == EIO_DT_DIR)
1788 ++dir;
1789 else if ((--oth)->type == EIO_DT_DIR)
1790 {
1791 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1792
1793 ++dir;
1794 }
1795 }
1796
1797 /* now sort the dirs only (dirs all have the same score) */
1798 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1799 }
1800
739 break; 1801 break;
1802 }
740 1803
741 name = entp->d_name; 1804 /* now add the entry to our list(s) */
1805 name = D_NAME (entp);
742 1806
1807 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1809 {
745 int len = strlen (name) + 1; 1810 int len = D_NAMLEN (entp) + 1;
746 1811
747 res++; 1812 while (ecb_expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1813 {
751 memlen *= 2; 1814 namesalloc *= 2;
752 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1815 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock);
755 1816
756 if (!names) 1817 if (!names)
757 break; 1818 break;
758 } 1819 }
759 1820
760 memcpy (names + memofs, name, len); 1821 memcpy (names + namesoffs, name, len);
1822
1823 if (dents)
1824 {
1825 struct eio_dirent *ent;
1826
1827 if (ecb_expect_false (dentoffs == dentalloc))
1828 {
1829 dentalloc *= 2;
1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1831
1832 if (!dents)
1833 break;
1834 }
1835
1836 ent = dents + dentoffs;
1837
1838 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1839 ent->namelen = len - 1;
1840 ent->inode = D_INO (entp);
1841
1842 inode_bits |= ent->inode;
1843
1844 switch (D_TYPE (entp))
1845 {
1846 default:
1847 ent->type = EIO_DT_UNKNOWN;
1848 flags |= EIO_READDIR_FOUND_UNKNOWN;
1849 break;
1850
1851 #ifdef DT_FIFO
1852 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1853 #endif
1854 #ifdef DT_CHR
1855 case DT_CHR: ent->type = EIO_DT_CHR; break;
1856 #endif
1857 #ifdef DT_MPC
1858 case DT_MPC: ent->type = EIO_DT_MPC; break;
1859 #endif
1860 #ifdef DT_DIR
1861 case DT_DIR: ent->type = EIO_DT_DIR; break;
1862 #endif
1863 #ifdef DT_NAM
1864 case DT_NAM: ent->type = EIO_DT_NAM; break;
1865 #endif
1866 #ifdef DT_BLK
1867 case DT_BLK: ent->type = EIO_DT_BLK; break;
1868 #endif
1869 #ifdef DT_MPB
1870 case DT_MPB: ent->type = EIO_DT_MPB; break;
1871 #endif
1872 #ifdef DT_REG
1873 case DT_REG: ent->type = EIO_DT_REG; break;
1874 #endif
1875 #ifdef DT_NWK
1876 case DT_NWK: ent->type = EIO_DT_NWK; break;
1877 #endif
1878 #ifdef DT_CMP
1879 case DT_CMP: ent->type = EIO_DT_CMP; break;
1880 #endif
1881 #ifdef DT_LNK
1882 case DT_LNK: ent->type = EIO_DT_LNK; break;
1883 #endif
1884 #ifdef DT_SOCK
1885 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1886 #endif
1887 #ifdef DT_DOOR
1888 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1889 #endif
1890 #ifdef DT_WHT
1891 case DT_WHT: ent->type = EIO_DT_WHT; break;
1892 #endif
1893 }
1894
1895 ent->score = 7;
1896
1897 if (flags & EIO_READDIR_DIRS_FIRST)
1898 {
1899 if (ent->type == EIO_DT_UNKNOWN)
1900 {
1901 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1902 ent->score = 1;
1903 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1904 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1905 }
1906 else if (ent->type == EIO_DT_DIR)
1907 ent->score = 0;
1908 }
1909 }
1910
761 memofs += len; 1911 namesoffs += len;
1912 ++dentoffs;
762 } 1913 }
1914
1915 if (EIO_CANCELLED (req))
1916 {
1917 errno = ECANCELED;
1918 break;
1919 }
1920
1921#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp))
1923 {
1924 FindClose (dirp);
1925 dirp = 0;
1926 }
1927#endif
763 } 1928 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769} 1929}
770 1930
771/*****************************************************************************/ 1931/*****************************************************************************/
772 1932
1933#define ALLOC(len) \
1934 if (!req->ptr2) \
1935 { \
1936 X_LOCK (wrklock); \
1937 req->flags |= EIO_FLAG_PTR2_FREE; \
1938 X_UNLOCK (wrklock); \
1939 req->ptr2 = malloc (len); \
1940 if (!req->ptr2) \
1941 { \
1942 errno = ENOMEM; \
1943 req->result = -1; \
1944 break; \
1945 } \
1946 }
1947
773X_THREAD_PROC (eio_proc) 1948X_THREAD_PROC (etp_proc)
774{ 1949{
775 eio_req *req; 1950 ETP_REQ *req;
776 struct timespec ts; 1951 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1952 etp_worker *self = (etp_worker *)thr_arg;
778 1953
1954#if HAVE_PRCTL_SET_NAME
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1956#endif
1957
779 /* try to distribute timeouts somewhat randomly */ 1958 /* try to distribute timeouts somewhat evenly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1960
782 for (;;) 1961 for (;;)
783 { 1962 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1963 ts.tv_sec = 0;
785 1964
786 X_LOCK (reqlock); 1965 X_LOCK (reqlock);
787 1966
788 for (;;) 1967 for (;;)
789 { 1968 {
790 self->req = req = reqq_shift (&req_queue); 1969 self->req = req = reqq_shift (&req_queue);
791 1970
792 if (req) 1971 if (req)
793 break; 1972 break;
794 1973
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1975 {
1976 X_UNLOCK (reqlock);
1977 X_LOCK (wrklock);
1978 --started;
1979 X_UNLOCK (wrklock);
1980 goto quit;
1981 }
1982
795 ++idle; 1983 ++idle;
796 1984
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1985 if (idle <= max_idle)
1986 /* we are allowed to idle, so do so without any timeout */
1987 X_COND_WAIT (reqwait, reqlock);
1988 else
798 { 1989 {
799 if (idle > max_idle) 1990 /* initialise timeout once */
800 { 1991 if (!ts.tv_sec)
801 --idle; 1992 ts.tv_sec = time (0) + idle_timeout;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808 1993
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1994 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
811 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1995 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
812 } 1996 }
813 1997
814 --idle; 1998 --idle;
815 } 1999 }
816 2000
817 --nready; 2001 --nready;
818 2002
819 X_UNLOCK (reqlock); 2003 X_UNLOCK (reqlock);
820 2004
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type) 2005 if (req->type < 0)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit; 2006 goto quit;
910 2007
911 default: 2008 ETP_EXECUTE (self, req);
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 2009
918 X_LOCK (reslock); 2010 X_LOCK (reslock);
919 2011
920 ++npending; 2012 ++npending;
921 2013
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 2014 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 2015 want_poll_cb ();
924 2016
925 self->req = 0; 2017 self->req = 0;
926 worker_clear (self); 2018 etp_worker_clear (self);
927 2019
928 X_UNLOCK (reslock); 2020 X_UNLOCK (reslock);
929 } 2021 }
930 2022
931quit: 2023quit:
932 X_LOCK (wrklock); 2024 X_LOCK (wrklock);
933 worker_free (self); 2025 etp_worker_free (self);
934 X_UNLOCK (wrklock); 2026 X_UNLOCK (wrklock);
935 2027
936 return 0; 2028 return 0;
937} 2029}
938 2030
939/*****************************************************************************/ 2031/*****************************************************************************/
940 2032
941static void eio_atfork_prepare (void) 2033int ecb_cold
2034eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 2035{
943 X_LOCK (wrklock);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE 2036#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952}
953
954static void eio_atfork_parent (void)
955{
956#if !HAVE_READDIR_R
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock); 2037 X_MUTEX_CREATE (preadwritelock);
961#endif 2038#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965}
966 2039
967static void eio_atfork_child (void) 2040 return etp_init (want_poll, done_poll);
968{ 2041}
969 eio_req *prv;
970 2042
971 while (prv = reqq_shift (&req_queue)) 2043ecb_inline void
972 eio_destroy (prv); 2044eio_api_destroy (eio_req *req)
2045{
2046 free (req);
2047}
973 2048
974 while (prv = reqq_shift (&res_queue)) 2049#define REQ(rtype) \
975 eio_destroy (prv); 2050 eio_req *req; \
2051 \
2052 req = (eio_req *)calloc (1, sizeof *req); \
2053 if (!req) \
2054 return 0; \
2055 \
2056 req->type = rtype; \
2057 req->pri = pri; \
2058 req->finish = cb; \
2059 req->data = data; \
2060 req->destroy = eio_api_destroy;
976 2061
977 while (wrk_first.next != &wrk_first) 2062#define SEND eio_submit (req); return req
2063
2064#define PATH \
2065 req->flags |= EIO_FLAG_PTR1_FREE; \
2066 req->ptr1 = strdup (path); \
2067 if (!req->ptr1) \
2068 { \
2069 eio_api_destroy (req); \
2070 return 0; \
2071 }
2072
2073static void
2074eio_execute (etp_worker *self, eio_req *req)
2075{
2076 if (ecb_expect_false (EIO_CANCELLED (req)))
978 { 2077 {
979 worker *wrk = wrk_first.next; 2078 req->result = -1;
980 2079 req->errorno = ECANCELED;
981 if (wrk->req) 2080 return;
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 } 2081 }
987 2082
988 started = 0; 2083 switch (req->type)
989 idle = 0; 2084 {
990 nreqs = 0; 2085 case EIO_READ: ALLOC (req->size);
991 nready = 0; 2086 req->result = req->offs >= 0
992 npending = 0; 2087 ? pread (req->int1, req->ptr2, req->size, req->offs)
2088 : read (req->int1, req->ptr2, req->size); break;
2089 case EIO_WRITE: req->result = req->offs >= 0
2090 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2091 : write (req->int1, req->ptr2, req->size); break;
993 2092
994 eio_atfork_parent (); 2093 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
995} 2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
996 2095
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
998{ 2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
999 want_poll_cb = want_poll; 2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1000 done_poll_cb = done_poll; 2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1001 2102
2103 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2104 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2105 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2106 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2107
2108 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2109 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2110 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2111 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2112 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2113 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2114
2115 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2116 case EIO_CLOSE: req->result = close (req->int1); break;
2117 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2118 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2119 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2120 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
2128 case EIO_READLINK: ALLOC (PATH_MAX);
2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2130
2131 case EIO_SYNC: req->result = 0; sync (); break;
2132 case EIO_FSYNC: req->result = fsync (req->int1); break;
2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2134 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2136 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2137 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2138 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2139 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2140 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2141
2142 case EIO_READDIR: eio__scandir (req, self); break;
2143
2144 case EIO_BUSY:
1002#ifdef _WIN32 2145#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 2146 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 2147#else
1005 X_MUTEX_CHECK (reqlock); 2148 {
1006 X_MUTEX_CHECK (reqwait); 2149 struct timeval tv;
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009 2150
1010 X_COND_CHECK (reqwait); 2151 tv.tv_sec = req->nv1;
1011#endif 2152 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012 2153
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2154 req->result = select (0, 0, 0, 0, &tv);
1014} 2155 }
2156#endif
2157 break;
1015 2158
1016#if 0 2159 case EIO_UTIME:
2160 case EIO_FUTIME:
2161 {
2162 struct timeval tv[2];
2163 struct timeval *times;
1017 2164
2165 if (req->nv1 != -1. || req->nv2 != -1.)
2166 {
2167 tv[0].tv_sec = req->nv1;
2168 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2169 tv[1].tv_sec = req->nv2;
2170 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2171
2172 times = tv;
2173 }
2174 else
2175 times = 0;
2176
2177 req->result = req->type == EIO_FUTIME
2178 ? futimes (req->int1, times)
2179 : utimes (req->ptr1, times);
2180 }
2181 break;
2182
2183 case EIO_GROUP:
2184 abort (); /* handled in eio_request */
2185
2186 case EIO_NOP:
2187 req->result = 0;
2188 break;
2189
2190 case EIO_CUSTOM:
2191 req->feed (req);
2192 break;
2193
2194 default:
2195 errno = ENOSYS;
2196 req->result = -1;
2197 break;
2198 }
2199
2200 req->errorno = errno;
2201}
2202
2203#ifndef EIO_NO_WRAPPERS
2204
2205eio_req *eio_nop (int pri, eio_cb cb, void *data)
2206{
2207 REQ (EIO_NOP); SEND;
2208}
2209
2210eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2211{
2212 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2213}
2214
1018eio_req *eio_fsync (int fd, eio_cb cb); 2215eio_req *eio_sync (int pri, eio_cb cb, void *data)
2216{
2217 REQ (EIO_SYNC); SEND;
2218}
2219
2220eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2221{
2222 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2223}
2224
2225eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2226{
2227 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2228}
2229
1019eio_req *eio_fdatasync (int fd, eio_cb cb); 2230eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020eio_req *eio_dupclose (int fd, eio_cb cb); 2231{
2232 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2233}
2234
2235eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2236{
2237 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2238}
2239
2240eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2243}
2244
2245eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2246{
2247 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2248}
2249
2250eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2251{
2252 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2253}
2254
2255eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2256{
2257 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2258}
2259
2260eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2261{
2262 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2263}
2264
2265eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2266{
2267 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2268}
2269
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2270eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2271{
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2272 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2273}
2274
2275eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2276{
2277 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2278}
2279
2280eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2281{
2282 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2283}
2284
2285eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2286{
2287 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2288}
2289
2290eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2291{
2292 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2293}
2294
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2295eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2296{
2297 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2298}
2299
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2300eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2301{
2302 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2303}
2304
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2305eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2306{
2307 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2308}
2309
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2310eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2311{
2312 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2313}
2314
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2315eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2316{
2317 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2318}
2319
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2320eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2321{
2322 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2323}
2324
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2325eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2326{
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2327 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2328}
2329
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2330eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2331{
2332 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2333}
2334
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2335eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2336{
2337 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2338}
2339
2340eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2341{
2342 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2343}
2344
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2345eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2346{
2347 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2348}
2349
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2350eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2351{
2352 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2353}
2354
2355static eio_req *
2356eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2357{
2358 REQ (type); PATH; SEND;
2359}
2360
2361eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2362{
2363 return eio__1path (EIO_READLINK, path, pri, cb, data);
2364}
2365
2366eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2367{
2368 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2369}
2370
2371eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2372{
2373 return eio__1path (EIO_STAT, path, pri, cb, data);
2374}
2375
2376eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2377{
2378 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2379}
2380
2381eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2382{
2383 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2384}
2385
1040eio_req *eio_unlink (const char *path, eio_cb cb); 2386eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2387{
2388 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2389}
2390
1041eio_req *eio_rmdir (const char *path, eio_cb cb); 2391eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2392{
2393 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2394}
2395
2396eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2397{
2398 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2399}
2400
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2401eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2402{
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2403 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2404}
2405
2406static eio_req *
2407eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2408{
2409 REQ (type); PATH;
2410
2411 req->flags |= EIO_FLAG_PTR2_FREE;
2412 req->ptr2 = strdup (new_path);
2413 if (!req->ptr2)
2414 {
2415 eio_api_destroy (req);
2416 return 0;
2417 }
2418
2419 SEND;
2420}
2421
2422eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2423{
2424 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2425}
2426
2427eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2428{
2429 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2430}
2431
2432eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2433{
2434 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2435}
2436
2437eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2438{
2439 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2440}
2441
2442#endif
2443
2444eio_req *eio_grp (eio_cb cb, void *data)
2445{
2446 const int pri = EIO_PRI_MAX;
2447
2448 REQ (EIO_GROUP); SEND;
2449}
2450
2451#undef REQ
2452#undef PATH
2453#undef SEND
2454
2455/*****************************************************************************/
2456/* grp functions */
2457
1046void 2458void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 {
1130 dREQ;
1131
1132 req->type = ix;
1133 req->sv1 = newSVsv (fh);
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141
1142 if (!SvREADONLY (data))
1143 {
1144 SvREADONLY_on (data);
1145 req->flags |= FLAG_SV2_RO_OFF;
1146 }
1147
1148 EIO_SEND;
1149 }
1150}
1151
1152void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159
1160 data = newSV (NAME_MAX);
1161 SvPOK_on (data);
1162
1163 req->type = EIO_READLINK;
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170}
1171
1172void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2459eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2460{
1421 grp->int2 = limit; 2461 grp->int2 = limit;
1422 grp->feed = feed; 2462 grp->feed = feed;
1423 2463
1424 grp_try_feed (grp); 2464 grp_try_feed (grp);
1425} 2465}
1426 2466
2467void
1427void eio_grp_limit (eio_req *grp, int limit) 2468eio_grp_limit (eio_req *grp, int limit)
1428{ 2469{
1429 grp->int2 = limit; 2470 grp->int2 = limit;
1430 2471
1431 grp_try_feed (grp); 2472 grp_try_feed (grp);
1432} 2473}
1433 2474
2475void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2476eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2477{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2478 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2479
2480 grp->flags |= EIO_FLAG_GROUPADD;
1437 2481
1438 ++grp->size; 2482 ++grp->size;
1439 req->grp = grp; 2483 req->grp = grp;
1440 2484
1441 req->grp_prev = 0; 2485 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2489 grp->grp_first->grp_prev = req;
1446 2490
1447 grp->grp_first = req; 2491 grp->grp_first = req;
1448} 2492}
1449 2493
2494/*****************************************************************************/
2495/* misc garbage */
1450 2496
2497eio_ssize_t
2498eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2499{
2500 return eio__sendfile (ofd, ifd, offset, count);
2501}
2502

Diff Legend

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