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

Comparing libeio/eio.c (file contents):
Revision 1.4 by root, Sun May 11 00:10:15 2008 UTC vs.
Revision 1.91 by root, Sun Jul 17 04:20:04 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
… …
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76static void eio_destroy (eio_req *req);
14 77
15#ifndef EIO_FINISH 78#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 80#endif
18 81
… …
22 85
23#ifndef EIO_FEED 86#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 88#endif
26 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
27#ifdef _WIN32 101#ifdef _WIN32
28 102
29 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
118
119 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
123 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
124 #define mknod(path,mode,dev) EIO_ENOSYS ()
125 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129
130 /* we could even stat and see if it exists */
131 static int
132 symlink (const char *old, const char *neu)
133 {
134 #if WINVER >= 0x0600
135 if (CreateSymbolicLink (neu, old, 1))
136 return 0;
137
138 if (CreateSymbolicLink (neu, old, 0))
139 return 0;
140 #endif
141
142 return EIO_ERRNO (ENOENT, -1);
143 }
144
145 /* POSIX API only */
146 #define CreateHardLink(neu,old,flags) 0
147 #define CreateSymbolicLink(neu,old,flags) 0
148
149 struct statvfs
150 {
151 int dummy;
152 };
153
154 #define DT_DIR EIO_DT_DIR
155 #define DT_REG EIO_DT_REG
156 #define D_NAME(entp) entp.cFileName
157 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 158
31#else 159#else
32 160
33# include "config.h"
34# include <sys/time.h> 161 #include <sys/time.h>
35# include <sys/select.h> 162 #include <sys/select.h>
163 #include <sys/statvfs.h>
36# include <unistd.h> 164 #include <unistd.h>
37# include <utime.h> 165 #include <utime.h>
38# include <signal.h> 166 #include <signal.h>
39# include <dirent.h> 167 #include <dirent.h>
40 168
169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
170 #include <sys/mman.h>
171 #endif
172
173 #define D_NAME(entp) entp->d_name
174
175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
177 #define _DIRENT_HAVE_D_TYPE /* sigh */
178 #define D_INO(de) (de)->d_fileno
179 #define D_NAMLEN(de) (de)->d_namlen
180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
181 #define D_INO(de) (de)->d_ino
182 #endif
183
184 #ifdef _D_EXACT_NAMLEN
185 #undef D_NAMLEN
186 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
187 #endif
188
189 #ifdef _DIRENT_HAVE_D_TYPE
190 #define D_TYPE(de) (de)->d_type
191 #endif
192
41# ifndef EIO_STRUCT_DIRENT 193 #ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 194 #define EIO_STRUCT_DIRENT struct dirent
195 #endif
196
43# endif 197#endif
44
45#endif
46
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50 198
51#if HAVE_SENDFILE 199#if HAVE_SENDFILE
52# if __linux 200# if __linux
53# include <sys/sendfile.h> 201# include <sys/sendfile.h>
54# elif __freebsd 202# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 203# include <sys/socket.h>
56# include <sys/uio.h> 204# include <sys/uio.h>
57# elif __hpux 205# elif __hpux
58# include <sys/socket.h> 206# include <sys/socket.h>
59# elif __solaris /* not yet */ 207# elif __solaris
60# include <sys/sendfile.h> 208# include <sys/sendfile.h>
61# else 209# else
62# error sendfile support requested but not available 210# error sendfile support requested but not available
63# endif 211# endif
64#endif 212#endif
65 213
66/* number of seconds after which an idle threads exit */ 214#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 215# define D_TYPE(de) 0
216#endif
217#ifndef D_INO
218# define D_INO(de) 0
219#endif
220#ifndef D_NAMLEN
221# define D_NAMLEN(entp) strlen (D_NAME (entp))
222#endif
68 223
69/* used for struct dirent, AIX doesn't provide it */ 224/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 225#ifndef NAME_MAX
71# define NAME_MAX 4096 226# define NAME_MAX 4096
72#endif 227#endif
73 228
229/* used for readlink etc. */
230#ifndef PATH_MAX
231# define PATH_MAX 4096
232#endif
233
74/* buffer size for various temporary buffers */ 234/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 235#define EIO_BUFSIZE 65536
76 236
77#define dBUF \ 237#define dBUF \
78 char *eio_buf; \ 238 char *eio_buf; \
79 X_LOCK (wrklock); \ 239 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 240 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 241 ETP_WORKER_UNLOCK (self); \
242 errno = ENOMEM; \
82 if (!eio_buf) \ 243 if (!eio_buf) \
83 return -1; 244 return -1;
84 245
85#define EIO_TICKS ((1000000 + 1023) >> 10) 246#define EIO_TICKS ((1000000 + 1023) >> 10)
86 247
248#define ETP_PRI_MIN EIO_PRI_MIN
249#define ETP_PRI_MAX EIO_PRI_MAX
250
251struct etp_worker;
252
253#define ETP_REQ eio_req
254#define ETP_DESTROY(req) eio_destroy (req)
255static int eio_finish (eio_req *req);
256#define ETP_FINISH(req) eio_finish (req)
257static void eio_execute (struct etp_worker *self, eio_req *req);
258#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
259
260#define ETP_WORKER_CLEAR(req) \
261 if (wrk->dbuf) \
262 { \
263 free (wrk->dbuf); \
264 wrk->dbuf = 0; \
265 }
266
267#define ETP_WORKER_COMMON \
268 void *dbuf;
269
270/*****************************************************************************/
271
272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
273
274/* calculate time difference in ~1/EIO_TICKS of a second */
275ecb_inline int
276tvdiff (struct timeval *tv1, struct timeval *tv2)
277{
278 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
279 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
280}
281
282static unsigned int started, idle, wanted = 4;
283
87static void (*want_poll_cb) (void); 284static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 285static void (*done_poll_cb) (void);
89 286
90static unsigned int max_poll_time = 0; 287static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 288static unsigned int max_poll_reqs; /* reslock */
92 289
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 290static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 291static volatile unsigned int nready; /* reqlock */
95{ 292static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 293static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 294static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 295
100static unsigned int started, idle, wanted; 296static xmutex_t wrklock;
297static xmutex_t reslock;
298static xmutex_t reqlock;
299static xcond_t reqwait;
300
301#if !HAVE_PREADWRITE
302/*
303 * make our pread/pwrite emulation safe against themselves, but not against
304 * normal read/write by using a mutex. slows down execution a lot,
305 * but that's your problem, not mine.
306 */
307static xmutex_t preadwritelock = X_MUTEX_INIT;
308#endif
309
310typedef struct etp_worker
311{
312 /* locked by wrklock */
313 struct etp_worker *prev, *next;
314
315 xthread_t tid;
316
317 /* locked by reslock, reqlock or wrklock */
318 ETP_REQ *req; /* currently processed request */
319
320 ETP_WORKER_COMMON
321} etp_worker;
322
323static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
324
325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 327
102/* worker threads management */ 328/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 329
105typedef struct worker { 330static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 331etp_worker_clear (etp_worker *wrk)
120{ 332{
121 if (wrk->dirp) 333 ETP_WORKER_CLEAR (wrk);
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126
127 if (wrk->dbuf)
128 {
129 free (wrk->dbuf);
130 wrk->dbuf = 0;
131 }
132} 334}
133 335
134static void worker_free (worker *wrk) 336static void ecb_cold
337etp_worker_free (etp_worker *wrk)
135{ 338{
136 wrk->next->prev = wrk->prev; 339 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 340 wrk->prev->next = wrk->next;
138 341
139 free (wrk); 342 free (wrk);
140} 343}
141 344
142static volatile unsigned int nreqs, nready, npending; 345static unsigned int
143static volatile unsigned int max_idle = 4; 346etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 347{
348 int retval;
349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
350 retval = nreqs;
351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 352 return retval;
152} 353}
153 354
154unsigned int eio_nready (void) 355static unsigned int
356etp_nready (void)
155{ 357{
156 unsigned int retval; 358 unsigned int retval;
157 359
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 361 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 363
162 return retval; 364 return retval;
163} 365}
164 366
165unsigned int eio_npending (void) 367static unsigned int
368etp_npending (void)
166{ 369{
167 unsigned int retval; 370 unsigned int retval;
168 371
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 373 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 375
173 return retval; 376 return retval;
174} 377}
175 378
176unsigned int eio_nthreads (void) 379static unsigned int
380etp_nthreads (void)
177{ 381{
178 unsigned int retval; 382 unsigned int retval;
179 383
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 385 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 392 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 393 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 394 * per shift, the most expensive operation.
191 */ 395 */
192typedef struct { 396typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 397 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 398 int size;
195} reqq; 399} etp_reqq;
196 400
197static reqq req_queue; 401static etp_reqq req_queue;
198static reqq res_queue; 402static etp_reqq res_queue;
199 403
200static int reqq_push (reqq *q, eio_req *req) 404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
415static int ecb_noinline
416reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 417{
202 int pri = req->pri; 418 int pri = req->pri;
203 req->next = 0; 419 req->next = 0;
204 420
205 if (q->qe[pri]) 421 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 427 q->qe[pri] = q->qs[pri] = req;
212 428
213 return q->size++; 429 return q->size++;
214} 430}
215 431
216static eio_req *reqq_shift (reqq *q) 432static ETP_REQ * ecb_noinline
433reqq_shift (etp_reqq *q)
217{ 434{
218 int pri; 435 int pri;
219 436
220 if (!q->size) 437 if (!q->size)
221 return 0; 438 return 0;
222 439
223 --q->size; 440 --q->size;
224 441
225 for (pri = EIO_NUM_PRI; pri--; ) 442 for (pri = ETP_NUM_PRI; pri--; )
226 { 443 {
227 eio_req *req = q->qs[pri]; 444 eio_req *req = q->qs[pri];
228 445
229 if (req) 446 if (req)
230 { 447 {
… …
236 } 453 }
237 454
238 abort (); 455 abort ();
239} 456}
240 457
241static void grp_try_feed (eio_req *grp) 458static void ecb_cold
459etp_thread_init (void)
242{ 460{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 461#if !HAVE_PREADWRITE
244 { 462 X_MUTEX_CREATE (preadwritelock);
245 int old_len = grp->size; 463#endif
246 464 X_MUTEX_CREATE (wrklock);
247 EIO_FEED (grp); 465 X_MUTEX_CREATE (reslock);
248 466 X_MUTEX_CREATE (reqlock);
249 /* stop if no progress has been made */ 467 X_COND_CREATE (reqwait);
250 if (old_len == grp->size)
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256} 468}
257 469
258static int eio_finish (eio_req *req); 470static void ecb_cold
259 471etp_atfork_child (void)
260static int grp_dec (eio_req *grp)
261{ 472{
262 --grp->size; 473 reqq_init (&req_queue);
474 reqq_init (&res_queue);
263 475
264 /* call feeder, if applicable */ 476 wrk_first.next =
265 grp_try_feed (grp); 477 wrk_first.prev = &wrk_first;
266 478
267 /* finish, if done */ 479 started = 0;
268 if (!grp->size && grp->int1) 480 idle = 0;
269 return eio_finish (grp); 481 nreqs = 0;
270 else 482 nready = 0;
483 npending = 0;
484
485 etp_thread_init ();
486}
487
488static void ecb_cold
489etp_once_init (void)
490{
491 etp_thread_init ();
492 X_THREAD_ATFORK (0, 0, etp_atfork_child);
493}
494
495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
497{
498 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
499
500 pthread_once (&doinit, etp_once_init);
501
502 want_poll_cb = want_poll;
503 done_poll_cb = done_poll;
504
271 return 0; 505 return 0;
272} 506}
273 507
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR2_FREE)
277 free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 req->flags |= EIO_FLAG_CANCELLED;
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 508X_THREAD_PROC (etp_proc);
323 509
324static void start_thread (void) 510static void ecb_cold
511etp_start_thread (void)
325{ 512{
326 worker *wrk = calloc (1, sizeof (worker)); 513 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 514
328 if (!wrk) 515 /*TODO*/
329 croak ("unable to allocate worker thread data"); 516 assert (("unable to allocate worker thread data", wrk));
330 517
331 X_LOCK (wrklock); 518 X_LOCK (wrklock);
332 519
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 520 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 521 {
335 wrk->prev = &wrk_first; 522 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 523 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 524 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 525 wrk_first.next = wrk;
… …
342 free (wrk); 529 free (wrk);
343 530
344 X_UNLOCK (wrklock); 531 X_UNLOCK (wrklock);
345} 532}
346 533
534static void
347static void maybe_start_thread (void) 535etp_maybe_start_thread (void)
348{ 536{
349 if (eio_nthreads () >= wanted) 537 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 538 return;
351 539
352 /* todo: maybe use idle here, but might be less exact */ 540 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 541 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 542 return;
355 543
356 start_thread (); 544 etp_start_thread ();
357} 545}
358 546
359void eio_submit (eio_req *req) 547static void ecb_cold
360{ 548etp_end_thread (void)
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void)
373{ 549{
374 eio_req *req = calloc (1, sizeof (eio_req)); 550 eio_req *req = calloc (1, sizeof (eio_req));
375 551
376 req->type = EIO_QUIT; 552 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 553 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 554
379 X_LOCK (reqlock); 555 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 556 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 557 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 558 X_UNLOCK (reqlock);
… …
384 X_LOCK (wrklock); 560 X_LOCK (wrklock);
385 --started; 561 --started;
386 X_UNLOCK (wrklock); 562 X_UNLOCK (wrklock);
387} 563}
388 564
389void eio_set_max_poll_time (double nseconds) 565static int
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 566etp_poll (void)
426{ 567{
427 int maxreqs = max_poll_reqs; 568 unsigned int maxreqs;
569 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 570 struct timeval tv_start, tv_now;
429 eio_req *req;
430 571
572 X_LOCK (reslock);
573 maxreqs = max_poll_reqs;
574 maxtime = max_poll_time;
575 X_UNLOCK (reslock);
576
431 if (max_poll_time) 577 if (maxtime)
432 gettimeofday (&tv_start, 0); 578 gettimeofday (&tv_start, 0);
433 579
434 for (;;) 580 for (;;)
435 { 581 {
582 ETP_REQ *req;
583
436 maybe_start_thread (); 584 etp_maybe_start_thread ();
437 585
438 X_LOCK (reslock); 586 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 587 req = reqq_shift (&res_queue);
440 588
441 if (req) 589 if (req)
… …
449 X_UNLOCK (reslock); 597 X_UNLOCK (reslock);
450 598
451 if (!req) 599 if (!req)
452 return 0; 600 return 0;
453 601
602 X_LOCK (reqlock);
454 --nreqs; 603 --nreqs;
604 X_UNLOCK (reqlock);
455 605
456 if (req->type == EIO_GROUP && req->size) 606 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457 { 607 {
458 req->int1 = 1; /* mark request as delayed */ 608 req->int1 = 1; /* mark request as delayed */
459 continue; 609 continue;
460 } 610 }
461 else 611 else
462 { 612 {
463 int res = eio_finish (req); 613 int res = ETP_FINISH (req);
464 if (res) 614 if (ecb_expect_false (res))
465 return res; 615 return res;
466 } 616 }
467 617
468 if (maxreqs && !--maxreqs) 618 if (ecb_expect_false (maxreqs && !--maxreqs))
469 break; 619 break;
470 620
471 if (max_poll_time) 621 if (maxtime)
472 { 622 {
473 gettimeofday (&tv_now, 0); 623 gettimeofday (&tv_now, 0);
474 624
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 625 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 626 break;
477 } 627 }
478 } 628 }
479 629
480 errno = EAGAIN; 630 errno = EAGAIN;
481 return -1; 631 return -1;
482} 632}
483 633
634static void
635etp_cancel (ETP_REQ *req)
636{
637 req->cancelled = 1;
638
639 eio_grp_cancel (req);
640}
641
642static void
643etp_submit (ETP_REQ *req)
644{
645 req->pri -= ETP_PRI_MIN;
646
647 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
648 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
649
650 if (ecb_expect_false (req->type == EIO_GROUP))
651 {
652 /* I hope this is worth it :/ */
653 X_LOCK (reqlock);
654 ++nreqs;
655 X_UNLOCK (reqlock);
656
657 X_LOCK (reslock);
658
659 ++npending;
660
661 if (!reqq_push (&res_queue, req) && want_poll_cb)
662 want_poll_cb ();
663
664 X_UNLOCK (reslock);
665 }
666 else
667 {
668 X_LOCK (reqlock);
669 ++nreqs;
670 ++nready;
671 reqq_push (&req_queue, req);
672 X_COND_SIGNAL (reqwait);
673 X_UNLOCK (reqlock);
674
675 etp_maybe_start_thread ();
676 }
677}
678
679static void ecb_cold
680etp_set_max_poll_time (double nseconds)
681{
682 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
683 max_poll_time = nseconds * EIO_TICKS;
684 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
685}
686
687static void ecb_cold
688etp_set_max_poll_reqs (unsigned int maxreqs)
689{
690 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
691 max_poll_reqs = maxreqs;
692 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
693}
694
695static void ecb_cold
696etp_set_max_idle (unsigned int nthreads)
697{
698 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
699 max_idle = nthreads;
700 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
701}
702
703static void ecb_cold
704etp_set_idle_timeout (unsigned int seconds)
705{
706 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
707 idle_timeout = seconds;
708 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
709}
710
711static void ecb_cold
712etp_set_min_parallel (unsigned int nthreads)
713{
714 if (wanted < nthreads)
715 wanted = nthreads;
716}
717
718static void ecb_cold
719etp_set_max_parallel (unsigned int nthreads)
720{
721 if (wanted > nthreads)
722 wanted = nthreads;
723
724 while (started > wanted)
725 etp_end_thread ();
726}
727
728/*****************************************************************************/
729
730static void
731grp_try_feed (eio_req *grp)
732{
733 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
734 {
735 grp->flags &= ~EIO_FLAG_GROUPADD;
736
737 EIO_FEED (grp);
738
739 /* stop if no progress has been made */
740 if (!(grp->flags & EIO_FLAG_GROUPADD))
741 {
742 grp->feed = 0;
743 break;
744 }
745 }
746}
747
748static int
749grp_dec (eio_req *grp)
750{
751 --grp->size;
752
753 /* call feeder, if applicable */
754 grp_try_feed (grp);
755
756 /* finish, if done */
757 if (!grp->size && grp->int1)
758 return eio_finish (grp);
759 else
760 return 0;
761}
762
763static void
764eio_destroy (eio_req *req)
765{
766 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
767 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
768
769 EIO_DESTROY (req);
770}
771
772static int
773eio_finish (eio_req *req)
774{
775 int res = EIO_FINISH (req);
776
777 if (req->grp)
778 {
779 int res2;
780 eio_req *grp = req->grp;
781
782 /* unlink request */
783 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
784 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
785
786 if (grp->grp_first == req)
787 grp->grp_first = req->grp_next;
788
789 res2 = grp_dec (grp);
790
791 if (!res)
792 res = res2;
793 }
794
795 eio_destroy (req);
796
797 return res;
798}
799
800void
801eio_grp_cancel (eio_req *grp)
802{
803 for (grp = grp->grp_first; grp; grp = grp->grp_next)
804 eio_cancel (grp);
805}
806
807void
808eio_cancel (eio_req *req)
809{
810 etp_cancel (req);
811}
812
813void
814eio_submit (eio_req *req)
815{
816 etp_submit (req);
817}
818
819unsigned int
820eio_nreqs (void)
821{
822 return etp_nreqs ();
823}
824
825unsigned int
826eio_nready (void)
827{
828 return etp_nready ();
829}
830
831unsigned int
832eio_npending (void)
833{
834 return etp_npending ();
835}
836
837unsigned int ecb_cold
838eio_nthreads (void)
839{
840 return etp_nthreads ();
841}
842
843void ecb_cold
844eio_set_max_poll_time (double nseconds)
845{
846 etp_set_max_poll_time (nseconds);
847}
848
849void ecb_cold
850eio_set_max_poll_reqs (unsigned int maxreqs)
851{
852 etp_set_max_poll_reqs (maxreqs);
853}
854
855void ecb_cold
856eio_set_max_idle (unsigned int nthreads)
857{
858 etp_set_max_idle (nthreads);
859}
860
861void ecb_cold
862eio_set_idle_timeout (unsigned int seconds)
863{
864 etp_set_idle_timeout (seconds);
865}
866
867void ecb_cold
868eio_set_min_parallel (unsigned int nthreads)
869{
870 etp_set_min_parallel (nthreads);
871}
872
873void ecb_cold
874eio_set_max_parallel (unsigned int nthreads)
875{
876 etp_set_max_parallel (nthreads);
877}
878
879int eio_poll (void)
880{
881 return etp_poll ();
882}
883
484/*****************************************************************************/ 884/*****************************************************************************/
485/* work around various missing functions */ 885/* work around various missing functions */
486 886
487#if !HAVE_PREADWRITE 887#if !HAVE_PREADWRITE
888# undef pread
889# undef pwrite
488# define pread aio_pread 890# define pread eio__pread
489# define pwrite aio_pwrite 891# define pwrite eio__pwrite
490 892
491/* 893static eio_ssize_t
492 * make our pread/pwrite safe against themselves, but not against 894eio__pread (int fd, void *buf, size_t count, off_t offset)
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499{ 895{
500 ssize_t res; 896 eio_ssize_t res;
501 off_t ooffset; 897 off_t ooffset;
502 898
503 X_LOCK (preadwritelock); 899 X_LOCK (preadwritelock);
504 ooffset = lseek (fd, 0, SEEK_CUR); 900 ooffset = lseek (fd, 0, SEEK_CUR);
505 lseek (fd, offset, SEEK_SET); 901 lseek (fd, offset, SEEK_SET);
… …
508 X_UNLOCK (preadwritelock); 904 X_UNLOCK (preadwritelock);
509 905
510 return res; 906 return res;
511} 907}
512 908
513static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 909static eio_ssize_t
910eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514{ 911{
515 ssize_t res; 912 eio_ssize_t res;
516 off_t ooffset; 913 off_t ooffset;
517 914
518 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
519 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
520 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
521 res = write (fd, buf, count); 918 res = write (fd, buf, count);
522 lseek (fd, offset, SEEK_SET); 919 lseek (fd, ooffset, SEEK_SET);
523 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
524 921
525 return res; 922 return res;
526} 923}
527#endif 924#endif
528 925
529#ifndef HAVE_FUTIMES 926#ifndef HAVE_UTIMES
530 927
928# undef utimes
531# define utimes(path,times) aio_utimes (path, times) 929# define utimes(path,times) eio__utimes (path, times)
532# define futimes(fd,times) aio_futimes (fd, times)
533 930
931static int
534static int aio_utimes (const char *filename, const struct timeval times[2]) 932eio__utimes (const char *filename, const struct timeval times[2])
535{ 933{
536 if (times) 934 if (times)
537 { 935 {
538 struct utimbuf buf; 936 struct utimbuf buf;
539 937
… …
544 } 942 }
545 else 943 else
546 return utime (filename, 0); 944 return utime (filename, 0);
547} 945}
548 946
947#endif
948
949#ifndef HAVE_FUTIMES
950
951# undef futimes
952# define futimes(fd,times) eio__futimes (fd, times)
953
954static int
549static int aio_futimes (int fd, const struct timeval tv[2]) 955eio__futimes (int fd, const struct timeval tv[2])
550{ 956{
551 errno = ENOSYS; 957 errno = ENOSYS;
552 return -1; 958 return -1;
553} 959}
554 960
555#endif 961#endif
556 962
557#if !HAVE_FDATASYNC 963#if !HAVE_FDATASYNC
964# undef fdatasync
558# define fdatasync fsync 965# define fdatasync(fd) fsync (fd)
559#endif 966#endif
967
968/* sync_file_range always needs emulation */
969static int
970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
971{
972#if HAVE_SYNC_FILE_RANGE
973 int res;
974
975 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
976 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
977 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
978 {
979 flags = 0
980 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
981 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
982 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
983 }
984
985 res = sync_file_range (fd, offset, nbytes, flags);
986
987 if (!res || errno != ENOSYS)
988 return res;
989#endif
990
991 /* even though we could play tricks with the flags, it's better to always
992 * call fdatasync, as that matches the expectation of its users best */
993 return fdatasync (fd);
994}
995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
1005}
560 1006
561#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
1008# undef readahead
562# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563 1010
1011static eio_ssize_t
564static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565{ 1013{
566 size_t todo = count; 1014 size_t todo = count;
567 dBUF; 1015 dBUF;
568 1016
569 while (todo > 0) 1017 while (todo > 0)
… …
579 return count; 1027 return count;
580} 1028}
581 1029
582#endif 1030#endif
583 1031
584#if !HAVE_READDIR_R
585# define readdir_r aio_readdir_r
586
587static mutex_t readdirlock = X_MUTEX_INIT;
588
589static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590{
591 X_DIRENT *e;
592 int errorno;
593
594 X_LOCK (readdirlock);
595
596 e = readdir (dirp);
597 errorno = errno;
598
599 if (e)
600 {
601 *res = ent;
602 strcpy (ent->d_name, e->d_name);
603 }
604 else
605 *res = 0;
606
607 X_UNLOCK (readdirlock);
608
609 errno = errorno;
610 return e ? 0 : -1;
611}
612#endif
613
614/* sendfile always needs emulation */ 1032/* sendfile always needs emulation */
1033static eio_ssize_t
615static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1034eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
616{ 1035{
1036 eio_ssize_t written = 0;
617 ssize_t res; 1037 eio_ssize_t res;
618 1038
619 if (!count) 1039 if (!count)
620 return 0; 1040 return 0;
621 1041
1042 for (;;)
1043 {
1044#ifdef __APPLE__
1045# undef HAVE_SENDFILE /* broken, as everything on os x */
1046#endif
622#if HAVE_SENDFILE 1047#if HAVE_SENDFILE
623# if __linux 1048# if __linux
1049 off_t soffset = offset;
624 res = sendfile (ofd, ifd, &offset, count); 1050 res = sendfile (ofd, ifd, &soffset, count);
625 1051
626# elif __freebsd 1052# elif __FreeBSD__
627 /* 1053 /*
628 * Of course, the freebsd sendfile is a dire hack with no thoughts 1054 * Of course, the freebsd sendfile is a dire hack with no thoughts
629 * wasted on making it similar to other I/O functions. 1055 * wasted on making it similar to other I/O functions.
630 */ 1056 */
631 {
632 off_t sbytes; 1057 off_t sbytes;
633 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1058 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634 1059
635 if (res < 0 && sbytes) 1060 #if 0 /* according to the manpage, this is correct, but broken behaviour */
636 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1061 /* freebsd' sendfile will return 0 on success */
1062 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1063 /* not on e.g. EIO or EPIPE - sounds broken */
1064 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637 res = sbytes; 1065 res = sbytes;
638 } 1066 #endif
1067
1068 /* according to source inspection, this is correct, and useful behaviour */
1069 if (sbytes)
1070 res = sbytes;
1071
1072# elif defined (__APPLE__)
1073 off_t sbytes = count;
1074 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1075
1076 /* according to the manpage, sbytes is always valid */
1077 if (sbytes)
1078 res = sbytes;
639 1079
640# elif __hpux 1080# elif __hpux
641 res = sendfile (ofd, ifd, offset, count, 0, 0); 1081 res = sendfile (ofd, ifd, offset, count, 0, 0);
642 1082
643# elif __solaris 1083# elif __solaris
644 {
645 struct sendfilevec vec; 1084 struct sendfilevec vec;
646 size_t sbytes; 1085 size_t sbytes;
647 1086
648 vec.sfv_fd = ifd; 1087 vec.sfv_fd = ifd;
649 vec.sfv_flag = 0; 1088 vec.sfv_flag = 0;
650 vec.sfv_off = offset; 1089 vec.sfv_off = offset;
651 vec.sfv_len = count; 1090 vec.sfv_len = count;
652 1091
653 res = sendfilev (ofd, &vec, 1, &sbytes); 1092 res = sendfilev (ofd, &vec, 1, &sbytes);
654 1093
655 if (res < 0 && sbytes) 1094 if (res < 0 && sbytes)
656 res = sbytes; 1095 res = sbytes;
657 }
658 1096
659# endif 1097# endif
1098
1099#elif defined (_WIN32) && 0
1100 /* does not work, just for documentation of what would need to be done */
1101 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1102 /* libeio guarantees that the file offset does not change, and windows */
1103 /* has no way to get an independent handle to the same file description */
1104 HANDLE h = TO_SOCKET (ifd);
1105 SetFilePointer (h, offset, 0, FILE_BEGIN);
1106 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1107
660#else 1108#else
661 res = -1; 1109 res = -1;
662 errno = ENOSYS; 1110 errno = ENOSYS;
663#endif 1111#endif
664 1112
1113 /* we assume sendfile can copy at least 128mb in one go */
1114 if (res <= 128 * 1024 * 1024)
1115 {
1116 if (res > 0)
1117 written += res;
1118
1119 if (written)
1120 return written;
1121
1122 break;
1123 }
1124 else
1125 {
1126 /* if we requested more, then probably the kernel was lazy */
1127 written += res;
1128 offset += res;
1129 count -= res;
1130
1131 if (!count)
1132 return written;
1133 }
1134 }
1135
665 if (res < 0 1136 if (res < 0
666 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1137 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1138 /* BSDs */
1139#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1140 || errno == ENOTSUP
1141#endif
1142#ifdef EOPNOTSUPP /* windows */
1143 || errno == EOPNOTSUPP /* BSDs */
1144#endif
667#if __solaris 1145#if __solaris
668 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1146 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
669#endif 1147#endif
670 ) 1148 )
671 ) 1149 )
… …
675 1153
676 res = 0; 1154 res = 0;
677 1155
678 while (count) 1156 while (count)
679 { 1157 {
680 ssize_t cnt; 1158 eio_ssize_t cnt;
681 1159
682 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1160 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
683 1161
684 if (cnt <= 0) 1162 if (cnt <= 0)
685 { 1163 {
… …
702 } 1180 }
703 1181
704 return res; 1182 return res;
705} 1183}
706 1184
1185#ifdef PAGESIZE
1186# define eio_pagesize() PAGESIZE
1187#else
1188static intptr_t
1189eio_pagesize (void)
1190{
1191 static intptr_t page;
1192
1193 if (!page)
1194 page = sysconf (_SC_PAGESIZE);
1195
1196 return page;
1197}
1198#endif
1199
1200static void
1201eio_page_align (void **addr, size_t *length)
1202{
1203 intptr_t mask = eio_pagesize () - 1;
1204
1205 /* round down addr */
1206 intptr_t adj = mask & (intptr_t)*addr;
1207
1208 *addr = (void *)((intptr_t)*addr - adj);
1209 *length += adj;
1210
1211 /* round up length */
1212 *length = (*length + mask) & ~mask;
1213}
1214
1215#if !_POSIX_MEMLOCK
1216# define eio__mlockall(a) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlockall (int flags)
1221{
1222 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1223 extern int mallopt (int, int);
1224 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1225 #endif
1226
1227 if (EIO_MCL_CURRENT != MCL_CURRENT
1228 || EIO_MCL_FUTURE != MCL_FUTURE)
1229 {
1230 flags = 0
1231 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1232 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1233 }
1234
1235 return mlockall (flags);
1236}
1237#endif
1238
1239#if !_POSIX_MEMLOCK_RANGE
1240# define eio__mlock(a,b) EIO_ENOSYS ()
1241#else
1242
1243static int
1244eio__mlock (void *addr, size_t length)
1245{
1246 eio_page_align (&addr, &length);
1247
1248 return mlock (addr, length);
1249}
1250
1251#endif
1252
1253#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1254# define eio__msync(a,b,c) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__msync (void *mem, size_t len, int flags)
1259{
1260 eio_page_align (&mem, &len);
1261
1262 if (EIO_MS_ASYNC != MS_SYNC
1263 || EIO_MS_INVALIDATE != MS_INVALIDATE
1264 || EIO_MS_SYNC != MS_SYNC)
1265 {
1266 flags = 0
1267 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1268 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1269 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1270 }
1271
1272 return msync (mem, len, flags);
1273}
1274
1275#endif
1276
1277static int
1278eio__mtouch (eio_req *req)
1279{
1280 void *mem = req->ptr2;
1281 size_t len = req->size;
1282 int flags = req->int1;
1283
1284 eio_page_align (&mem, &len);
1285
1286 {
1287 intptr_t addr = (intptr_t)mem;
1288 intptr_t end = addr + len;
1289 intptr_t page = eio_pagesize ();
1290
1291 if (addr < end)
1292 if (flags & EIO_MT_MODIFY) /* modify */
1293 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1294 else
1295 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1296 }
1297
1298 return 0;
1299}
1300
1301/*****************************************************************************/
1302/* requests implemented outside eio_execute, because they are so large */
1303
1304static void
1305eio__realpath (eio_req *req, etp_worker *self)
1306{
1307 char *rel = req->ptr1;
1308 char *res;
1309 char *tmp1, *tmp2;
1310#if SYMLOOP_MAX > 32
1311 int symlinks = SYMLOOP_MAX;
1312#else
1313 int symlinks = 32;
1314#endif
1315
1316 req->result = -1;
1317
1318 errno = EINVAL;
1319 if (!rel)
1320 return;
1321
1322 errno = ENOENT;
1323 if (!*rel)
1324 return;
1325
1326 if (!req->ptr2)
1327 {
1328 X_LOCK (wrklock);
1329 req->flags |= EIO_FLAG_PTR2_FREE;
1330 X_UNLOCK (wrklock);
1331 req->ptr2 = malloc (PATH_MAX * 3);
1332
1333 errno = ENOMEM;
1334 if (!req->ptr2)
1335 return;
1336 }
1337
1338 res = req->ptr2;
1339 tmp1 = res + PATH_MAX;
1340 tmp2 = tmp1 + PATH_MAX;
1341
1342#if 0 /* disabled, the musl way to do things is just too racy */
1343#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1344 /* on linux we may be able to ask the kernel */
1345 {
1346 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1347
1348 if (fd >= 0)
1349 {
1350 sprintf (tmp1, "/proc/self/fd/%d", fd);
1351 req->result = readlink (tmp1, res, PATH_MAX);
1352 close (fd);
1353
1354 /* here we should probably stat the open file and the disk file, to make sure they still match */
1355
1356 if (req->result > 0)
1357 goto done;
1358 }
1359 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1360 return;
1361 }
1362#endif
1363#endif
1364
1365 if (*rel != '/')
1366 {
1367 if (!getcwd (res, PATH_MAX))
1368 return;
1369
1370 if (res [1]) /* only use if not / */
1371 res += strlen (res);
1372 }
1373
1374 while (*rel)
1375 {
1376 eio_ssize_t len, linklen;
1377 char *beg = rel;
1378
1379 while (*rel && *rel != '/')
1380 ++rel;
1381
1382 len = rel - beg;
1383
1384 if (!len) /* skip slashes */
1385 {
1386 ++rel;
1387 continue;
1388 }
1389
1390 if (beg [0] == '.')
1391 {
1392 if (len == 1)
1393 continue; /* . - nop */
1394
1395 if (beg [1] == '.' && len == 2)
1396 {
1397 /* .. - back up one component, if possible */
1398
1399 while (res != req->ptr2)
1400 if (*--res == '/')
1401 break;
1402
1403 continue;
1404 }
1405 }
1406
1407 errno = ENAMETOOLONG;
1408 if (res + 1 + len + 1 >= tmp1)
1409 return;
1410
1411 /* copy one component */
1412 *res = '/';
1413 memcpy (res + 1, beg, len);
1414
1415 /* zero-terminate, for readlink */
1416 res [len + 1] = 0;
1417
1418 /* now check if it's a symlink */
1419 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1420
1421 if (linklen < 0)
1422 {
1423 if (errno != EINVAL)
1424 return;
1425
1426 /* it's a normal directory. hopefully */
1427 res += len + 1;
1428 }
1429 else
1430 {
1431 /* yay, it was a symlink - build new path in tmp2 */
1432 int rellen = strlen (rel);
1433
1434 errno = ENAMETOOLONG;
1435 if (linklen + 1 + rellen >= PATH_MAX)
1436 return;
1437
1438 errno = ELOOP;
1439 if (!--symlinks)
1440 return;
1441
1442 if (*tmp1 == '/')
1443 res = req->ptr2; /* symlink resolves to an absolute path */
1444
1445 /* we need to be careful, as rel might point into tmp2 already */
1446 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1447 tmp2 [linklen] = '/';
1448 memcpy (tmp2, tmp1, linklen);
1449
1450 rel = tmp2;
1451 }
1452 }
1453
1454 /* special case for the lone root path */
1455 if (res == req->ptr2)
1456 *res++ = '/';
1457
1458 req->result = res - (char *)req->ptr2;
1459
1460done:
1461 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1462}
1463
1464static signed char
1465eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1466{
1467 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1468 : a->inode < b->inode ? -1
1469 : a->inode > b->inode ? 1
1470 : 0;
1471}
1472
1473#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1474
1475#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1476#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1477
1478static void
1479eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1480{
1481 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1482 unsigned char *bit = bits;
1483
1484 assert (CHAR_BIT == 8);
1485 assert (sizeof (eio_dirent) * 8 < 256);
1486 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1487 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1488
1489 if (size <= EIO_SORT_FAST)
1490 return;
1491
1492 /* first prepare an array of bits to test in our radix sort */
1493 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1494 /* inode_bits must contain all inodes ORed together */
1495 /* which is used to skip bits that are 0 everywhere, which is very common */
1496 {
1497 eio_ino_t endianness;
1498 int i, j;
1499
1500 /* we store the byte offset of byte n into byte n of "endianness" */
1501 for (i = 0; i < sizeof (eio_ino_t); ++i)
1502 ((unsigned char *)&endianness)[i] = i;
1503
1504 *bit++ = 0;
1505
1506 for (i = 0; i < sizeof (eio_ino_t); ++i)
1507 {
1508 /* shifting off the byte offsets out of "endianness" */
1509 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1510 endianness >>= 8;
1511
1512 for (j = 0; j < 8; ++j)
1513 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1514 *bit++ = offs + j;
1515 }
1516
1517 for (j = 0; j < 8; ++j)
1518 if (score_bits & (1 << j))
1519 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1520 }
1521
1522 /* now actually do the sorting (a variant of MSD radix sort) */
1523 {
1524 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1525 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1526 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1527 int stk_idx = 0;
1528
1529 base_stk [stk_idx] = dents;
1530 end_stk [stk_idx] = dents + size;
1531 bit_stk [stk_idx] = bit - 1;
1532
1533 do
1534 {
1535 base = base_stk [stk_idx];
1536 end = end_stk [stk_idx];
1537 bit = bit_stk [stk_idx];
1538
1539 for (;;)
1540 {
1541 unsigned char O = *bit >> 3;
1542 unsigned char M = 1 << (*bit & 7);
1543
1544 eio_dirent *a = base;
1545 eio_dirent *b = end;
1546
1547 if (b - a < EIO_SORT_CUTOFF)
1548 break;
1549
1550 /* now bit-partition the array on the bit */
1551 /* this ugly asymmetric loop seems to perform much better than typical */
1552 /* partition algos found in the literature */
1553 do
1554 if (!(((unsigned char *)a)[O] & M))
1555 ++a;
1556 else if (!(((unsigned char *)--b)[O] & M))
1557 {
1558 eio_dirent tmp = *a; *a = *b; *b = tmp;
1559 ++a;
1560 }
1561 while (b > a);
1562
1563 /* next bit, or stop, if no bits left in this path */
1564 if (!*--bit)
1565 break;
1566
1567 base_stk [stk_idx] = a;
1568 end_stk [stk_idx] = end;
1569 bit_stk [stk_idx] = bit;
1570 ++stk_idx;
1571
1572 end = a;
1573 }
1574 }
1575 while (stk_idx--);
1576 }
1577}
1578
1579static void
1580eio_dent_insertion_sort (eio_dirent *dents, int size)
1581{
1582 /* first move the smallest element to the front, to act as a sentinel */
1583 {
1584 int i;
1585 eio_dirent *min = dents;
1586
1587 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1588 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1589 if (EIO_DENT_CMP (dents [i], <, *min))
1590 min = &dents [i];
1591
1592 /* swap elements 0 and j (minimum) */
1593 {
1594 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1595 }
1596 }
1597
1598 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1599 {
1600 eio_dirent *i, *j;
1601
1602 for (i = dents + 1; i < dents + size; ++i)
1603 {
1604 eio_dirent value = *i;
1605
1606 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1607 j [1] = j [0];
1608
1609 j [1] = value;
1610 }
1611 }
1612}
1613
1614static void
1615eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1616{
1617 if (size <= 1)
1618 return; /* our insertion sort relies on size > 0 */
1619
1620 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1621 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1622 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1623
1624 /* use an insertion sort at the end, or for small arrays, */
1625 /* as insertion sort is more efficient for small partitions */
1626 eio_dent_insertion_sort (dents, size);
1627}
1628
707/* read a full directory */ 1629/* read a full directory */
1630static void
708static void scandir_ (eio_req *req, worker *self) 1631eio__scandir (eio_req *req, etp_worker *self)
709{ 1632{
1633 char *name, *names;
1634 int namesalloc = 4096 - sizeof (void *) * 4;
1635 int namesoffs = 0;
1636 int flags = req->int1;
1637 eio_dirent *dents = 0;
1638 int dentalloc = 128;
1639 int dentoffs = 0;
1640 eio_ino_t inode_bits = 0;
1641#ifdef _WIN32
1642 HANDLE dirp;
1643 WIN32_FIND_DATA entp;
1644#else
710 DIR *dirp; 1645 DIR *dirp;
711 union
712 {
713 EIO_STRUCT_DIRENT d;
714 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715 } *u;
716 EIO_STRUCT_DIRENT *entp; 1646 EIO_STRUCT_DIRENT *entp;
717 char *name, *names; 1647#endif
718 int memlen = 4096;
719 int memofs = 0;
720 int res = 0;
721 1648
722 X_LOCK (wrklock); 1649 req->result = -1;
1650
1651 if (!(flags & EIO_READDIR_DENTS))
1652 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1653
1654#ifdef _WIN32
1655 {
1656 char *path = malloc (MAX_PATH);
1657 _snprintf (path, MAX_PATH, "%s/*", (const char *)req->ptr1);
1658 dirp = FindFirstFile (path, &entp);
1659 free (path);
1660
1661 if (!dirp)
1662 {
1663 switch (GetLastError ())
1664 {
1665 case ERROR_FILE_NOT_FOUND:
1666 req->result = 0;
1667 break;
1668
1669 case ERROR_PATH_NOT_FOUND:
1670 case ERROR_NO_MORE_FILES:
1671 errno = ENOENT;
1672 break;
1673
1674 case ERROR_NOT_ENOUGH_MEMORY:
1675 errno = ENOMEM;
1676 break;
1677
1678 default:
1679 errno = EINVAL;
1680 break;
1681 }
1682
1683 }
1684 }
1685#else
723 self->dirp = dirp = opendir (req->ptr1); 1686 dirp = opendir (req->ptr1);
724 self->dbuf = u = malloc (sizeof (*u)); 1687#endif
1688
725 req->flags |= EIO_FLAG_PTR2_FREE; 1689 if (req->flags & EIO_FLAG_PTR1_FREE)
1690 free (req->ptr1);
1691
1692 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1693 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726 req->ptr2 = names = malloc (memlen); 1694 req->ptr2 = names = malloc (namesalloc);
727 X_UNLOCK (wrklock);
728 1695
729 if (dirp && u && names) 1696 if (dirp && names && (!flags || dents))
730 for (;;) 1697 for (;;)
731 { 1698 {
1699 int more;
1700
1701#ifdef _WIN32
1702 more = dirp;
1703#else
732 errno = 0; 1704 errno = 0;
733 readdir_r (dirp, &u->d, &entp); 1705 entp = readdir (dirp);
1706 more = entp;
1707#endif
734 1708
735 if (!entp) 1709 if (!more)
1710 {
1711#ifndef _WIN32
1712 int old_errno = errno;
1713 closedir (dirp);
1714 errno = old_errno;
1715
1716 if (errno)
1717 break;
1718#endif
1719
1720 /* sort etc. */
1721 req->int1 = flags;
1722 req->result = dentoffs;
1723
1724 if (flags & EIO_READDIR_STAT_ORDER)
1725 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1726 else if (flags & EIO_READDIR_DIRS_FIRST)
1727 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1728 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1729 else
1730 {
1731 /* in this case, all is known, and we just put dirs first and sort them */
1732 eio_dirent *oth = dents + dentoffs;
1733 eio_dirent *dir = dents;
1734
1735 /* now partition dirs to the front, and non-dirs to the back */
1736 /* by walking from both sides and swapping if necessary */
1737 while (oth > dir)
1738 {
1739 if (dir->type == EIO_DT_DIR)
1740 ++dir;
1741 else if ((--oth)->type == EIO_DT_DIR)
1742 {
1743 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1744
1745 ++dir;
1746 }
1747 }
1748
1749 /* now sort the dirs only (dirs all have the same score) */
1750 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1751 }
1752
736 break; 1753 break;
1754 }
737 1755
738 name = entp->d_name; 1756 /* now add the entry to our list(s) */
1757 name = D_NAME (entp);
739 1758
1759 /* skip . and .. entries */
740 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1760 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741 { 1761 {
742 int len = strlen (name) + 1; 1762 int len = D_NAMLEN (entp) + 1;
743 1763
744 res++; 1764 while (ecb_expect_false (namesoffs + len > namesalloc))
745
746 while (memofs + len > memlen)
747 { 1765 {
748 memlen *= 2; 1766 namesalloc *= 2;
749 X_LOCK (wrklock);
750 req->ptr2 = names = realloc (names, memlen); 1767 req->ptr2 = names = realloc (names, namesalloc);
751 X_UNLOCK (wrklock);
752 1768
753 if (!names) 1769 if (!names)
754 break; 1770 break;
755 } 1771 }
756 1772
757 memcpy (names + memofs, name, len); 1773 memcpy (names + namesoffs, name, len);
1774
1775 if (dents)
1776 {
1777 struct eio_dirent *ent;
1778
1779 if (ecb_expect_false (dentoffs == dentalloc))
1780 {
1781 dentalloc *= 2;
1782 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1783
1784 if (!dents)
1785 break;
1786 }
1787
1788 ent = dents + dentoffs;
1789
1790 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1791 ent->namelen = len - 1;
1792 ent->inode = D_INO (entp);
1793
1794 inode_bits |= ent->inode;
1795
1796 switch (D_TYPE (entp))
1797 {
1798 default:
1799 ent->type = EIO_DT_UNKNOWN;
1800 flags |= EIO_READDIR_FOUND_UNKNOWN;
1801 break;
1802
1803 #ifdef DT_FIFO
1804 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1805 #endif
1806 #ifdef DT_CHR
1807 case DT_CHR: ent->type = EIO_DT_CHR; break;
1808 #endif
1809 #ifdef DT_MPC
1810 case DT_MPC: ent->type = EIO_DT_MPC; break;
1811 #endif
1812 #ifdef DT_DIR
1813 case DT_DIR: ent->type = EIO_DT_DIR; break;
1814 #endif
1815 #ifdef DT_NAM
1816 case DT_NAM: ent->type = EIO_DT_NAM; break;
1817 #endif
1818 #ifdef DT_BLK
1819 case DT_BLK: ent->type = EIO_DT_BLK; break;
1820 #endif
1821 #ifdef DT_MPB
1822 case DT_MPB: ent->type = EIO_DT_MPB; break;
1823 #endif
1824 #ifdef DT_REG
1825 case DT_REG: ent->type = EIO_DT_REG; break;
1826 #endif
1827 #ifdef DT_NWK
1828 case DT_NWK: ent->type = EIO_DT_NWK; break;
1829 #endif
1830 #ifdef DT_CMP
1831 case DT_CMP: ent->type = EIO_DT_CMP; break;
1832 #endif
1833 #ifdef DT_LNK
1834 case DT_LNK: ent->type = EIO_DT_LNK; break;
1835 #endif
1836 #ifdef DT_SOCK
1837 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1838 #endif
1839 #ifdef DT_DOOR
1840 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1841 #endif
1842 #ifdef DT_WHT
1843 case DT_WHT: ent->type = EIO_DT_WHT; break;
1844 #endif
1845 }
1846
1847 ent->score = 7;
1848
1849 if (flags & EIO_READDIR_DIRS_FIRST)
1850 {
1851 if (ent->type == EIO_DT_UNKNOWN)
1852 {
1853 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1854 ent->score = 1;
1855 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1856 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1857 }
1858 else if (ent->type == EIO_DT_DIR)
1859 ent->score = 0;
1860 }
1861 }
1862
758 memofs += len; 1863 namesoffs += len;
1864 ++dentoffs;
759 } 1865 }
1866
1867 if (EIO_CANCELLED (req))
1868 {
1869 errno = ECANCELED;
1870 break;
1871 }
1872
1873#ifdef _WIN32
1874 if (!FindNextFile (dirp, &entp))
1875 {
1876 FindClose (dirp);
1877 dirp = 0;
1878 }
1879#endif
760 } 1880 }
761
762 if (errno)
763 res = -1;
764
765 req->result = res;
766} 1881}
767 1882
768/*****************************************************************************/ 1883/*****************************************************************************/
769 1884
1885#define ALLOC(len) \
1886 if (!req->ptr2) \
1887 { \
1888 X_LOCK (wrklock); \
1889 req->flags |= EIO_FLAG_PTR2_FREE; \
1890 X_UNLOCK (wrklock); \
1891 req->ptr2 = malloc (len); \
1892 if (!req->ptr2) \
1893 { \
1894 errno = ENOMEM; \
1895 req->result = -1; \
1896 break; \
1897 } \
1898 }
1899
770X_THREAD_PROC (eio_proc) 1900X_THREAD_PROC (etp_proc)
771{ 1901{
772 eio_req *req; 1902 ETP_REQ *req;
773 struct timespec ts; 1903 struct timespec ts;
774 worker *self = (worker *)thr_arg; 1904 etp_worker *self = (etp_worker *)thr_arg;
775 1905
776 /* try to distribute timeouts somewhat randomly */ 1906 /* try to distribute timeouts somewhat evenly */
777 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1907 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778 1908
779 for (;;) 1909 for (;;)
780 { 1910 {
781 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1911 ts.tv_sec = 0;
782 1912
783 X_LOCK (reqlock); 1913 X_LOCK (reqlock);
784 1914
785 for (;;) 1915 for (;;)
786 { 1916 {
787 self->req = req = reqq_shift (&req_queue); 1917 self->req = req = reqq_shift (&req_queue);
788 1918
789 if (req) 1919 if (req)
790 break; 1920 break;
791 1921
1922 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1923 {
1924 X_UNLOCK (reqlock);
1925 X_LOCK (wrklock);
1926 --started;
1927 X_UNLOCK (wrklock);
1928 goto quit;
1929 }
1930
792 ++idle; 1931 ++idle;
793 1932
794 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1933 if (idle <= max_idle)
1934 /* we are allowed to idle, so do so without any timeout */
1935 X_COND_WAIT (reqwait, reqlock);
1936 else
795 { 1937 {
796 if (idle > max_idle) 1938 /* initialise timeout once */
797 { 1939 if (!ts.tv_sec)
798 --idle; 1940 ts.tv_sec = time (0) + idle_timeout;
799 X_UNLOCK (reqlock);
800 X_LOCK (wrklock);
801 --started;
802 X_UNLOCK (wrklock);
803 goto quit;
804 }
805 1941
806 /* we are allowed to idle, so do so without any timeout */
807 X_COND_WAIT (reqwait, reqlock); 1942 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1943 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
809 } 1944 }
810 1945
811 --idle; 1946 --idle;
812 } 1947 }
813 1948
814 --nready; 1949 --nready;
815 1950
816 X_UNLOCK (reqlock); 1951 X_UNLOCK (reqlock);
817 1952
818 errno = 0; /* strictly unnecessary */
819
820 if (!EIO_CANCELLED (req))
821 switch (req->type) 1953 if (req->type < 0)
822 {
823 case EIO_READ: req->result = req->offs >= 0
824 ? pread (req->int1, req->ptr2, req->size, req->offs)
825 : read (req->int1, req->ptr2, req->size); break;
826 case EIO_WRITE: req->result = req->offs >= 0
827 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
828 : write (req->int1, req->ptr2, req->size); break;
829
830 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845 case EIO_CLOSE: req->result = close (req->int1); break;
846 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856 case EIO_SYNC: req->result = 0; sync (); break;
857 case EIO_FSYNC: req->result = fsync (req->int1); break;
858 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860 case EIO_READDIR: scandir_ (req, self); break;
861
862 case EIO_BUSY:
863#ifdef _WIN32
864 Sleep (req->nv1 * 1000.);
865#else
866 {
867 struct timeval tv;
868
869 tv.tv_sec = req->nv1;
870 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872 req->result = select (0, 0, 0, 0, &tv);
873 }
874#endif
875 break;
876
877 case EIO_UTIME:
878 case EIO_FUTIME:
879 {
880 struct timeval tv[2];
881 struct timeval *times;
882
883 if (req->nv1 != -1. || req->nv2 != -1.)
884 {
885 tv[0].tv_sec = req->nv1;
886 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887 tv[1].tv_sec = req->nv2;
888 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890 times = tv;
891 }
892 else
893 times = 0;
894
895
896 req->result = req->type == EIO_FUTIME
897 ? futimes (req->int1, times)
898 : utimes (req->ptr1, times);
899 }
900
901 case EIO_GROUP:
902 case EIO_NOP:
903 break;
904
905 case EIO_QUIT:
906 goto quit; 1954 goto quit;
907 1955
908 default: 1956 ETP_EXECUTE (self, req);
909 req->result = -1;
910 break;
911 }
912
913 req->errorno = errno;
914 1957
915 X_LOCK (reslock); 1958 X_LOCK (reslock);
916 1959
917 ++npending; 1960 ++npending;
918 1961
919 if (!reqq_push (&res_queue, req) && want_poll_cb) 1962 if (!reqq_push (&res_queue, req) && want_poll_cb)
920 want_poll_cb (); 1963 want_poll_cb ();
921 1964
922 self->req = 0; 1965 self->req = 0;
923 worker_clear (self); 1966 etp_worker_clear (self);
924 1967
925 X_UNLOCK (reslock); 1968 X_UNLOCK (reslock);
926 } 1969 }
927 1970
928quit: 1971quit:
929 X_LOCK (wrklock); 1972 X_LOCK (wrklock);
930 worker_free (self); 1973 etp_worker_free (self);
931 X_UNLOCK (wrklock); 1974 X_UNLOCK (wrklock);
932 1975
933 return 0; 1976 return 0;
934} 1977}
935 1978
936/*****************************************************************************/ 1979/*****************************************************************************/
937 1980
938static void eio_atfork_prepare (void) 1981int ecb_cold
1982eio_init (void (*want_poll)(void), void (*done_poll)(void))
939{ 1983{
940 X_LOCK (wrklock); 1984 return etp_init (want_poll, done_poll);
941 X_LOCK (reqlock);
942 X_LOCK (reslock);
943#if !HAVE_PREADWRITE
944 X_LOCK (preadwritelock);
945#endif
946#if !HAVE_READDIR_R
947 X_LOCK (readdirlock);
948#endif
949} 1985}
950 1986
951static void eio_atfork_parent (void) 1987ecb_inline void
1988eio_api_destroy (eio_req *req)
952{ 1989{
953#if !HAVE_READDIR_R 1990 free (req);
954 X_UNLOCK (readdirlock);
955#endif
956#if !HAVE_PREADWRITE
957 X_UNLOCK (preadwritelock);
958#endif
959 X_UNLOCK (reslock);
960 X_UNLOCK (reqlock);
961 X_UNLOCK (wrklock);
962} 1991}
963 1992
964static void eio_atfork_child (void) 1993#define REQ(rtype) \
965{ 1994 eio_req *req; \
966 eio_req *prv; 1995 \
1996 req = (eio_req *)calloc (1, sizeof *req); \
1997 if (!req) \
1998 return 0; \
1999 \
2000 req->type = rtype; \
2001 req->pri = pri; \
2002 req->finish = cb; \
2003 req->data = data; \
2004 req->destroy = eio_api_destroy;
967 2005
968 while (prv = reqq_shift (&req_queue)) 2006#define SEND eio_submit (req); return req
969 eio_destroy (prv);
970 2007
971 while (prv = reqq_shift (&res_queue)) 2008#define PATH \
972 eio_destroy (prv); 2009 req->flags |= EIO_FLAG_PTR1_FREE; \
2010 req->ptr1 = strdup (path); \
2011 if (!req->ptr1) \
2012 { \
2013 eio_api_destroy (req); \
2014 return 0; \
2015 }
973 2016
974 while (wrk_first.next != &wrk_first) 2017static void
2018eio_execute (etp_worker *self, eio_req *req)
2019{
2020 if (ecb_expect_false (EIO_CANCELLED (req)))
975 { 2021 {
976 worker *wrk = wrk_first.next; 2022 req->result = -1;
2023 req->errorno = ECANCELED;
2024 return;
2025 }
977 2026
978 if (wrk->req) 2027 switch (req->type)
979 eio_destroy (wrk->req);
980
981 worker_clear (wrk);
982 worker_free (wrk);
983 } 2028 {
2029 case EIO_READ: ALLOC (req->size);
2030 req->result = req->offs >= 0
2031 ? pread (req->int1, req->ptr2, req->size, req->offs)
2032 : read (req->int1, req->ptr2, req->size); break;
2033 case EIO_WRITE: req->result = req->offs >= 0
2034 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2035 : write (req->int1, req->ptr2, req->size); break;
984 2036
985 started = 0; 2037 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
986 idle = 0; 2038 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
987 nreqs = 0;
988 nready = 0;
989 npending = 0;
990 2039
991 eio_atfork_parent (); 2040 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
992} 2041 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2042 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2043 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2044 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2045 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
993 2046
994int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2047 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
995{ 2048 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
996 want_poll_cb = want_poll; 2049 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
997 done_poll_cb = done_poll; 2050 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
998 2051
2052 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2053 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2054 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2055 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2056 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2057 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2058
2059 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2060 case EIO_CLOSE: req->result = close (req->int1); break;
2061 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2062 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2063 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2064 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2065 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2066 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2067 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2068 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2069
2070 case EIO_REALPATH: eio__realpath (req, self); break;
2071
2072 case EIO_READLINK: ALLOC (PATH_MAX);
2073 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2074
2075 case EIO_SYNC: req->result = 0; sync (); break;
2076 case EIO_FSYNC: req->result = fsync (req->int1); break;
2077 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2078 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2079 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2080 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2081 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2082 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2083 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2084
2085 case EIO_READDIR: eio__scandir (req, self); break;
2086
2087 case EIO_BUSY:
999#ifdef _WIN32 2088#ifdef _WIN32
1000 X_MUTEX_CHECK (wrklock); 2089 Sleep (req->nv1 * 1e3);
1001 X_MUTEX_CHECK (reslock); 2090#else
1002 X_MUTEX_CHECK (reqlock); 2091 {
1003 X_MUTEX_CHECK (reqwait); 2092 struct timeval tv;
1004 X_MUTEX_CHECK (preadwritelock);
1005 X_MUTEX_CHECK (readdirlock);
1006 2093
1007 X_COND_CHECK (reqwait); 2094 tv.tv_sec = req->nv1;
1008#endif 2095 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1009 2096
1010 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2097 req->result = select (0, 0, 0, 0, &tv);
1011} 2098 }
2099#endif
2100 break;
1012 2101
1013#if 0 2102 case EIO_UTIME:
2103 case EIO_FUTIME:
2104 {
2105 struct timeval tv[2];
2106 struct timeval *times;
1014 2107
2108 if (req->nv1 != -1. || req->nv2 != -1.)
2109 {
2110 tv[0].tv_sec = req->nv1;
2111 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2112 tv[1].tv_sec = req->nv2;
2113 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2114
2115 times = tv;
2116 }
2117 else
2118 times = 0;
2119
2120 req->result = req->type == EIO_FUTIME
2121 ? futimes (req->int1, times)
2122 : utimes (req->ptr1, times);
2123 }
2124 break;
2125
2126 case EIO_GROUP:
2127 abort (); /* handled in eio_request */
2128
2129 case EIO_NOP:
2130 req->result = 0;
2131 break;
2132
2133 case EIO_CUSTOM:
2134 req->feed (req);
2135 break;
2136
2137 default:
2138 errno = ENOSYS;
2139 req->result = -1;
2140 break;
2141 }
2142
2143 req->errorno = errno;
2144}
2145
2146#ifndef EIO_NO_WRAPPERS
2147
2148eio_req *eio_nop (int pri, eio_cb cb, void *data)
2149{
2150 REQ (EIO_NOP); SEND;
2151}
2152
2153eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2154{
2155 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2156}
2157
1015eio_req *eio_fsync (int fd, eio_cb cb); 2158eio_req *eio_sync (int pri, eio_cb cb, void *data)
2159{
2160 REQ (EIO_SYNC); SEND;
2161}
2162
2163eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2164{
2165 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2166}
2167
2168eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2169{
2170 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2171}
2172
2173eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2174{
2175 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2176}
2177
2178eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2179{
2180 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2181}
2182
2183eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2184{
2185 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2186}
2187
2188eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2189{
2190 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2191}
2192
2193eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2194{
2195 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2196}
2197
1016eio_req *eio_fdatasync (int fd, eio_cb cb); 2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017eio_req *eio_dupclose (int fd, eio_cb cb); 2199{
2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2201}
2202
2203eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2204{
2205 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2206}
2207
1018eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2208eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2209{
1020eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2210 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2211}
2212
2213eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2214{
2215 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2216}
2217
2218eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2219{
2220 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2221}
2222
2223eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2224{
2225 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2226}
2227
2228eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2231}
2232
1022eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2233eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2236}
2237
1023eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2238eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2241}
2242
1024eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2243eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2246}
2247
1025eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2248eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2249{
2250 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2251}
2252
1026eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2253eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2254{
2255 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2256}
2257
1027eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2258eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2259{
2260 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2261}
2262
1028eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2263eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2264{
1030eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2265 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2266}
2267
1032eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2268eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2269{
2270 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2271}
2272
1033eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2273eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2274{
2275 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2276}
2277
2278eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2279{
2280 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2281}
2282
1034eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2283eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2284{
2285 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2286}
2287
1035eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2288eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1036eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2289{
2290 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2291}
2292
2293static eio_req *
2294eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2295{
2296 REQ (type); PATH; SEND;
2297}
2298
2299eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2300{
2301 return eio__1path (EIO_READLINK, path, pri, cb, data);
2302}
2303
2304eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2305{
2306 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2307}
2308
2309eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2310{
2311 return eio__1path (EIO_STAT, path, pri, cb, data);
2312}
2313
2314eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2315{
2316 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2317}
2318
2319eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2320{
2321 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2322}
2323
1037eio_req *eio_unlink (const char *path, eio_cb cb); 2324eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2325{
2326 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2327}
2328
1038eio_req *eio_rmdir (const char *path, eio_cb cb); 2329eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2330{
2331 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2332}
2333
2334eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2335{
2336 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2337}
2338
1040eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2339eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2340{
1042eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2341 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2342}
2343
2344static eio_req *
2345eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2346{
2347 REQ (type); PATH;
2348
2349 req->flags |= EIO_FLAG_PTR2_FREE;
2350 req->ptr2 = strdup (new_path);
2351 if (!req->ptr2)
2352 {
2353 eio_api_destroy (req);
2354 return 0;
2355 }
2356
2357 SEND;
2358}
2359
2360eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2361{
2362 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2363}
2364
2365eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2366{
2367 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2368}
2369
2370eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2371{
2372 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2373}
2374
2375eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2376{
2377 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2378}
2379
2380#endif
2381
2382eio_req *eio_grp (eio_cb cb, void *data)
2383{
2384 const int pri = EIO_PRI_MAX;
2385
2386 REQ (EIO_GROUP); SEND;
2387}
2388
2389#undef REQ
2390#undef PATH
2391#undef SEND
2392
2393/*****************************************************************************/
2394/* grp functions */
2395
1043void 2396void
1044aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045 PROTOTYPE: $$$;$
1046 PPCODE:
1047{
1048 dREQ;
1049
1050 req->type = EIO_OPEN;
1051 req->sv1 = newSVsv (pathname);
1052 req->ptr1 = SvPVbyte_nolen (req->sv1);
1053 req->int1 = flags;
1054 req->int2 = mode;
1055
1056 EIO_SEND;
1057}
1058
1059void
1060aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061 PROTOTYPE: $;$
1062 ALIAS:
1063 aio_fsync = EIO_FSYNC
1064 aio_fdatasync = EIO_FDATASYNC
1065 PPCODE:
1066{
1067 dREQ;
1068
1069 req->type = ix;
1070 req->sv1 = newSVsv (fh);
1071 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073 EIO_SEND (req);
1074}
1075
1076void
1077aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078 PROTOTYPE: $;$
1079 PPCODE:
1080{
1081 dREQ;
1082
1083 req->type = EIO_CLOSE;
1084 req->sv1 = newSVsv (fh);
1085 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087 EIO_SEND (req);
1088}
1089
1090void
1091aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092 ALIAS:
1093 aio_read = EIO_READ
1094 aio_write = EIO_WRITE
1095 PROTOTYPE: $$$$$;$
1096 PPCODE:
1097{
1098 STRLEN svlen;
1099 char *svptr = SvPVbyte (data, svlen);
1100 UV len = SvUV (length);
1101
1102 SvUPGRADE (data, SVt_PV);
1103 SvPOK_on (data);
1104
1105 if (dataoffset < 0)
1106 dataoffset += svlen;
1107
1108 if (dataoffset < 0 || dataoffset > svlen)
1109 croak ("dataoffset outside of data scalar");
1110
1111 if (ix == EIO_WRITE)
1112 {
1113 /* write: check length and adjust. */
1114 if (!SvOK (length) || len + dataoffset > svlen)
1115 len = svlen - dataoffset;
1116 }
1117 else
1118 {
1119 /* read: grow scalar as necessary */
1120 svptr = SvGROW (data, len + dataoffset + 1);
1121 }
1122
1123 if (len < 0)
1124 croak ("length must not be negative");
1125
1126 {
1127 dREQ;
1128
1129 req->type = ix;
1130 req->sv1 = newSVsv (fh);
1131 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132 : IoOFP (sv_2io (fh)));
1133 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134 req->size = len;
1135 req->sv2 = SvREFCNT_inc (data);
1136 req->ptr2 = (char *)svptr + dataoffset;
1137 req->stroffset = dataoffset;
1138
1139 if (!SvREADONLY (data))
1140 {
1141 SvREADONLY_on (data);
1142 req->flags |= FLAG_SV2_RO_OFF;
1143 }
1144
1145 EIO_SEND;
1146 }
1147}
1148
1149void
1150aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151 PROTOTYPE: $$;$
1152 PPCODE:
1153{
1154 SV *data;
1155 dREQ;
1156
1157 data = newSV (NAME_MAX);
1158 SvPOK_on (data);
1159
1160 req->type = EIO_READLINK;
1161 req->sv1 = newSVsv (path);
1162 req->ptr1 = SvPVbyte_nolen (req->sv1);
1163 req->sv2 = data;
1164 req->ptr2 = SvPVbyte_nolen (data);
1165
1166 EIO_SEND;
1167}
1168
1169void
1170aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171 PROTOTYPE: $$$$;$
1172 PPCODE:
1173{
1174 dREQ;
1175
1176 req->type = EIO_SENDFILE;
1177 req->sv1 = newSVsv (out_fh);
1178 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179 req->sv2 = newSVsv (in_fh);
1180 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181 req->offs = SvVAL64 (in_offset);
1182 req->size = length;
1183
1184 EIO_SEND;
1185}
1186
1187void
1188aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189 PROTOTYPE: $$$;$
1190 PPCODE:
1191{
1192 dREQ;
1193
1194 req->type = EIO_READAHEAD;
1195 req->sv1 = newSVsv (fh);
1196 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197 req->offs = SvVAL64 (offset);
1198 req->size = length;
1199
1200 EIO_SEND;
1201}
1202
1203void
1204aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205 ALIAS:
1206 aio_stat = EIO_STAT
1207 aio_lstat = EIO_LSTAT
1208 PPCODE:
1209{
1210 dREQ;
1211
1212 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213 if (!req->ptr2)
1214 {
1215 req_destroy (req);
1216 croak ("out of memory during aio_stat statdata allocation");
1217 }
1218
1219 req->flags |= FLAG_PTR2_FREE;
1220 req->sv1 = newSVsv (fh_or_path);
1221
1222 if (SvPOK (fh_or_path))
1223 {
1224 req->type = ix;
1225 req->ptr1 = SvPVbyte_nolen (req->sv1);
1226 }
1227 else
1228 {
1229 req->type = EIO_FSTAT;
1230 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231 }
1232
1233 EIO_SEND;
1234}
1235
1236void
1237aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238 PPCODE:
1239{
1240 dREQ;
1241
1242 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244 req->sv1 = newSVsv (fh_or_path);
1245
1246 if (SvPOK (fh_or_path))
1247 {
1248 req->type = EIO_UTIME;
1249 req->ptr1 = SvPVbyte_nolen (req->sv1);
1250 }
1251 else
1252 {
1253 req->type = EIO_FUTIME;
1254 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255 }
1256
1257 EIO_SEND;
1258}
1259
1260void
1261aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262 PPCODE:
1263{
1264 dREQ;
1265
1266 req->sv1 = newSVsv (fh_or_path);
1267 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269 if (SvPOK (fh_or_path))
1270 {
1271 req->type = EIO_TRUNCATE;
1272 req->ptr1 = SvPVbyte_nolen (req->sv1);
1273 }
1274 else
1275 {
1276 req->type = EIO_FTRUNCATE;
1277 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278 }
1279
1280 EIO_SEND;
1281}
1282
1283void
1284aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285 ALIAS:
1286 aio_chmod = EIO_CHMOD
1287 aio_fchmod = EIO_FCHMOD
1288 aio_mkdir = EIO_MKDIR
1289 PPCODE:
1290{
1291 dREQ;
1292
1293 req->type = type;
1294 req->int2 = mode;
1295 req->sv1 = newSVsv (fh_or_path);
1296
1297 if (ix == EIO_FCHMOD)
1298 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299 else
1300 req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302 EIO_SEND;
1303}
1304
1305void
1306aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307 PPCODE:
1308{
1309 dREQ;
1310
1311 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313 req->sv1 = newSVsv (fh_or_path);
1314
1315 if (SvPOK (fh_or_path))
1316 {
1317 req->type = EIO_CHOWN;
1318 req->ptr1 = SvPVbyte_nolen (req->sv1);
1319 }
1320 else
1321 {
1322 req->type = EIO_FCHOWN;
1323 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324 }
1325
1326 EIO_SEND;
1327}
1328
1329void
1330aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331 ALIAS:
1332 aio_unlink = EIO_UNLINK
1333 aio_rmdir = EIO_RMDIR
1334 aio_readdir = EIO_READDIR
1335 PPCODE:
1336{
1337 dREQ;
1338
1339 req->type = ix;
1340 req->sv1 = newSVsv (pathname);
1341 req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343 EIO_SEND;
1344}
1345
1346void
1347aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348 ALIAS:
1349 aio_link = EIO_LINK
1350 aio_symlink = EIO_SYMLINK
1351 aio_rename = EIO_RENAME
1352 PPCODE:
1353{
1354 dREQ;
1355
1356 req->type = ix;
1357 req->sv1 = newSVsv (oldpath);
1358 req->ptr1 = SvPVbyte_nolen (req->sv1);
1359 req->sv2 = newSVsv (newpath);
1360 req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362 EIO_SEND;
1363}
1364
1365void
1366aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367 PPCODE:
1368{
1369 dREQ;
1370
1371 req->type = EIO_MKNOD;
1372 req->sv1 = newSVsv (pathname);
1373 req->ptr1 = SvPVbyte_nolen (req->sv1);
1374 req->int2 = (mode_t)mode;
1375 req->offs = dev;
1376
1377 EIO_SEND;
1378}
1379
1380void
1381aio_busy (double delay, SV *callback=&PL_sv_undef)
1382 PPCODE:
1383{
1384 dREQ;
1385
1386 req->type = EIO_BUSY;
1387 req->nv1 = delay < 0. ? 0. : delay;
1388
1389 EIO_SEND;
1390}
1391
1392void
1393aio_group (SV *callback=&PL_sv_undef)
1394 PROTOTYPE: ;$
1395 PPCODE:
1396{
1397 dREQ;
1398
1399 req->type = EIO_GROUP;
1400
1401 req_send (req);
1402 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403}
1404
1405void
1406aio_nop (SV *callback=&PL_sv_undef)
1407 ALIAS:
1408 aio_nop = EIO_NOP
1409 aio_sync = EIO_SYNC
1410 PPCODE:
1411{
1412 dREQ;
1413
1414#endif
1415
1416void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2397eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417{ 2398{
1418 grp->int2 = limit; 2399 grp->int2 = limit;
1419 grp->feed = feed; 2400 grp->feed = feed;
1420 2401
1421 grp_try_feed (grp); 2402 grp_try_feed (grp);
1422} 2403}
1423 2404
2405void
1424void eio_grp_limit (eio_req *grp, int limit) 2406eio_grp_limit (eio_req *grp, int limit)
1425{ 2407{
1426 grp->int2 = limit; 2408 grp->int2 = limit;
1427 2409
1428 grp_try_feed (grp); 2410 grp_try_feed (grp);
1429} 2411}
1430 2412
2413void
1431void eio_grp_add (eio_req *grp, eio_req *req) 2414eio_grp_add (eio_req *grp, eio_req *req)
1432{ 2415{
1433 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2416 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2417
2418 grp->flags |= EIO_FLAG_GROUPADD;
1434 2419
1435 ++grp->size; 2420 ++grp->size;
1436 req->grp = grp; 2421 req->grp = grp;
1437 2422
1438 req->grp_prev = 0; 2423 req->grp_prev = 0;
… …
1442 grp->grp_first->grp_prev = req; 2427 grp->grp_first->grp_prev = req;
1443 2428
1444 grp->grp_first = req; 2429 grp->grp_first = req;
1445} 2430}
1446 2431
2432/*****************************************************************************/
2433/* misc garbage */
1447 2434
2435eio_ssize_t
2436eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2437{
2438 etp_worker wrk;
2439 eio_ssize_t ret;
2440
2441 wrk.dbuf = 0;
2442
2443 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2444
2445 if (wrk.dbuf)
2446 free (wrk.dbuf);
2447
2448 return ret;
2449}
2450

Diff Legend

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