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

Diff Legend

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