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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.87 by root, Thu Jul 14 19:19:22 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>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63#include <sys/statvfs.h>
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
75
15#ifndef EIO_FINISH 76#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 78#endif
18 79
19#ifndef EIO_DESTROY 80#ifndef EIO_DESTROY
22 83
23#ifndef EIO_FEED 84#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 85# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 86#endif
26 87
88#ifndef EIO_FD_TO_WIN32_HANDLE
89# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
90#endif
91#ifndef EIO_WIN32_HANDLE_TO_FD
92# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
93#endif
94
95#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
96
97#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98
27#ifdef _WIN32 99#ifdef _WIN32
28 100
29 /*doh*/ 101 #define PAGESIZE 4096 /* GetSystemInfo? */
102
103 #ifdef EIO_STRUCT_STATI64
104 #define stat(path,buf) _stati64 (path,buf)
105 #define fstat(fd,buf) _fstati64 (path,buf)
106 #endif
107 #define lstat(path,buf) stat (path,buf)
108 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
109 #define mkdir(path,mode) _mkdir (path)
110 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
111
112 #define chown(path,uid,gid) EIO_ENOSYS ()
113 #define fchown(fd,uid,gid) EIO_ENOSYS ()
114 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
115 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
116 #define mknod(path,mode,dev) EIO_ENOSYS ()
117 #define sync() EIO_ENOSYS ()
118
119 /* we could even stat and see if it exists */
120 static int
121 symlink (const char *old, const char *neu)
122 {
123 if (CreateSymbolicLink (neu, old, 1))
124 return 0;
125
126 if (CreateSymbolicLink (neu, old, 0))
127 return 0;
128
129 return EIO_ERRNO (ENOENT, -1);
130 }
30 131
31#else 132#else
32 133
33# include "config.h"
34# include <sys/time.h> 134 #include <sys/time.h>
35# include <sys/select.h> 135 #include <sys/select.h>
136 #include <sys/statvfs.h>
36# include <unistd.h> 137 #include <unistd.h>
37# include <utime.h> 138 #include <utime.h>
38# include <signal.h> 139 #include <signal.h>
39# include <dirent.h> 140 #include <dirent.h>
40 141
142 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
143 #include <sys/mman.h>
144 #endif
145
146 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
147 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
148 #define _DIRENT_HAVE_D_TYPE /* sigh */
149 #define D_INO(de) (de)->d_fileno
150 #define D_NAMLEN(de) (de)->d_namlen
151 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
152 #define D_INO(de) (de)->d_ino
153 #endif
154
155 #ifdef _D_EXACT_NAMLEN
156 #undef D_NAMLEN
157 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
158 #endif
159
160 #ifdef _DIRENT_HAVE_D_TYPE
161 #define D_TYPE(de) (de)->d_type
162 #endif
163
41# ifndef EIO_STRUCT_DIRENT 164 #ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 165 #define EIO_STRUCT_DIRENT struct dirent
166 #endif
167
43# endif 168#endif
44
45#endif
46
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50 169
51#if HAVE_SENDFILE 170#if HAVE_SENDFILE
52# if __linux 171# if __linux
53# include <sys/sendfile.h> 172# include <sys/sendfile.h>
54# elif __freebsd 173# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 174# include <sys/socket.h>
56# include <sys/uio.h> 175# include <sys/uio.h>
57# elif __hpux 176# elif __hpux
58# include <sys/socket.h> 177# include <sys/socket.h>
59# elif __solaris /* not yet */ 178# elif __solaris
60# include <sys/sendfile.h> 179# include <sys/sendfile.h>
61# else 180# else
62# error sendfile support requested but not available 181# error sendfile support requested but not available
63# endif 182# endif
64#endif 183#endif
65 184
66/* number of seconds after which an idle threads exit */ 185#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 186# define D_TYPE(de) 0
187#endif
188#ifndef D_INO
189# define D_INO(de) 0
190#endif
191#ifndef D_NAMLEN
192# define D_NAMLEN(de) strlen ((de)->d_name)
193#endif
68 194
69/* used for struct dirent, AIX doesn't provide it */ 195/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 196#ifndef NAME_MAX
71# define NAME_MAX 4096 197# define NAME_MAX 4096
72#endif 198#endif
73 199
200/* used for readlink etc. */
201#ifndef PATH_MAX
202# define PATH_MAX 4096
203#endif
204
74/* buffer size for various temporary buffers */ 205/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 206#define EIO_BUFSIZE 65536
76 207
77#define dBUF \ 208#define dBUF \
78 char *eio_buf; \ 209 char *eio_buf; \
79 X_LOCK (wrklock); \ 210 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 211 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 212 ETP_WORKER_UNLOCK (self); \
213 errno = ENOMEM; \
82 if (!eio_buf) \ 214 if (!eio_buf) \
83 return -1; 215 return -1;
84 216
85#define EIO_TICKS ((1000000 + 1023) >> 10) 217#define EIO_TICKS ((1000000 + 1023) >> 10)
86 218
219#define ETP_PRI_MIN EIO_PRI_MIN
220#define ETP_PRI_MAX EIO_PRI_MAX
221
222struct etp_worker;
223
224#define ETP_REQ eio_req
225#define ETP_DESTROY(req) eio_destroy (req)
226static int eio_finish (eio_req *req);
227#define ETP_FINISH(req) eio_finish (req)
228static void eio_execute (struct etp_worker *self, eio_req *req);
229#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
230
231#define ETP_WORKER_CLEAR(req) \
232 if (wrk->dbuf) \
233 { \
234 free (wrk->dbuf); \
235 wrk->dbuf = 0; \
236 } \
237 \
238 if (wrk->dirp) \
239 { \
240 closedir (wrk->dirp); \
241 wrk->dirp = 0; \
242 }
243
244#define ETP_WORKER_COMMON \
245 void *dbuf; \
246 DIR *dirp;
247
248/*****************************************************************************/
249
250#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
251
252/* calculate time difference in ~1/EIO_TICKS of a second */
253ecb_inline int
254tvdiff (struct timeval *tv1, struct timeval *tv2)
255{
256 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
257 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
258}
259
260static unsigned int started, idle, wanted = 4;
261
87static void (*want_poll_cb) (void); 262static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 263static void (*done_poll_cb) (void);
89 264
90static unsigned int max_poll_time = 0; 265static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 266static unsigned int max_poll_reqs; /* reslock */
92 267
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 268static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 269static volatile unsigned int nready; /* reqlock */
95{ 270static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 271static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 272static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 273
100static unsigned int started, idle, wanted; 274static xmutex_t wrklock;
275static xmutex_t reslock;
276static xmutex_t reqlock;
277static xcond_t reqwait;
278
279#if !HAVE_PREADWRITE
280/*
281 * make our pread/pwrite emulation safe against themselves, but not against
282 * normal read/write by using a mutex. slows down execution a lot,
283 * but that's your problem, not mine.
284 */
285static xmutex_t preadwritelock = X_MUTEX_INIT;
286#endif
287
288typedef struct etp_worker
289{
290 /* locked by wrklock */
291 struct etp_worker *prev, *next;
292
293 xthread_t tid;
294
295 /* locked by reslock, reqlock or wrklock */
296 ETP_REQ *req; /* currently processed request */
297
298 ETP_WORKER_COMMON
299} etp_worker;
300
301static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
302
303#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
304#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 305
102/* worker threads management */ 306/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 307
105typedef struct worker { 308static 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) 309etp_worker_clear (etp_worker *wrk)
120{ 310{
121 if (wrk->dirp) 311 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} 312}
133 313
134static void worker_free (worker *wrk) 314static void ecb_cold
315etp_worker_free (etp_worker *wrk)
135{ 316{
136 wrk->next->prev = wrk->prev; 317 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 318 wrk->prev->next = wrk->next;
138 319
139 free (wrk); 320 free (wrk);
140} 321}
141 322
142static volatile unsigned int nreqs, nready, npending; 323static unsigned int
143static volatile unsigned int max_idle = 4; 324etp_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{ 325{
326 int retval;
327 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
328 retval = nreqs;
329 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 330 return retval;
152} 331}
153 332
154unsigned int eio_nready (void) 333static unsigned int
334etp_nready (void)
155{ 335{
156 unsigned int retval; 336 unsigned int retval;
157 337
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 338 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 339 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 340 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 341
162 return retval; 342 return retval;
163} 343}
164 344
165unsigned int eio_npending (void) 345static unsigned int
346etp_npending (void)
166{ 347{
167 unsigned int retval; 348 unsigned int retval;
168 349
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 350 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 351 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 352 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 353
173 return retval; 354 return retval;
174} 355}
175 356
176unsigned int eio_nthreads (void) 357static unsigned int
358etp_nthreads (void)
177{ 359{
178 unsigned int retval; 360 unsigned int retval;
179 361
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 363 retval = started;
188 * a somewhat faster data structure might be nice, but 370 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 371 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 372 * per shift, the most expensive operation.
191 */ 373 */
192typedef struct { 374typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 375 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 376 int size;
195} reqq; 377} etp_reqq;
196 378
197static reqq req_queue; 379static etp_reqq req_queue;
198static reqq res_queue; 380static etp_reqq res_queue;
199 381
200static int reqq_push (reqq *q, eio_req *req) 382static int ecb_noinline
383reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 384{
202 int pri = req->pri; 385 int pri = req->pri;
203 req->next = 0; 386 req->next = 0;
204 387
205 if (q->qe[pri]) 388 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 394 q->qe[pri] = q->qs[pri] = req;
212 395
213 return q->size++; 396 return q->size++;
214} 397}
215 398
216static eio_req *reqq_shift (reqq *q) 399static ETP_REQ * ecb_noinline
400reqq_shift (etp_reqq *q)
217{ 401{
218 int pri; 402 int pri;
219 403
220 if (!q->size) 404 if (!q->size)
221 return 0; 405 return 0;
222 406
223 --q->size; 407 --q->size;
224 408
225 for (pri = EIO_NUM_PRI; pri--; ) 409 for (pri = ETP_NUM_PRI; pri--; )
226 { 410 {
227 eio_req *req = q->qs[pri]; 411 eio_req *req = q->qs[pri];
228 412
229 if (req) 413 if (req)
230 { 414 {
236 } 420 }
237 421
238 abort (); 422 abort ();
239} 423}
240 424
241static void grp_try_feed (eio_req *grp) 425static void ecb_cold
426etp_thread_init (void)
242{ 427{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 428#if !HAVE_PREADWRITE
429 X_MUTEX_CREATE (preadwritelock);
430#endif
431 X_MUTEX_CREATE (wrklock);
432 X_MUTEX_CREATE (reslock);
433 X_MUTEX_CREATE (reqlock);
434 X_COND_CREATE (reqwait);
435}
436
437static void ecb_cold
438etp_atfork_prepare (void)
439{
440}
441
442static void ecb_cold
443etp_atfork_parent (void)
444{
445}
446
447static void ecb_cold
448etp_atfork_child (void)
449{
450 ETP_REQ *prv;
451
452 while ((prv = reqq_shift (&req_queue)))
453 ETP_DESTROY (prv);
454
455 while ((prv = reqq_shift (&res_queue)))
456 ETP_DESTROY (prv);
457
458 while (wrk_first.next != &wrk_first)
244 { 459 {
245 int old_len = grp->size; 460 etp_worker *wrk = wrk_first.next;
246 461
247 EIO_FEED (grp); 462 if (wrk->req)
463 ETP_DESTROY (wrk->req);
248 464
249 /* stop if no progress has been made */ 465 etp_worker_clear (wrk);
250 if (old_len == grp->size) 466 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 467 }
256}
257 468
258static int eio_finish (eio_req *req); 469 started = 0;
470 idle = 0;
471 nreqs = 0;
472 nready = 0;
473 npending = 0;
259 474
260static int grp_dec (eio_req *grp) 475 etp_thread_init ();
261{ 476}
262 --grp->size;
263 477
264 /* call feeder, if applicable */ 478static void ecb_cold
265 grp_try_feed (grp); 479etp_once_init (void)
480{
481 etp_thread_init ();
482 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
483}
266 484
267 /* finish, if done */ 485static int ecb_cold
268 if (!grp->size && grp->int1) 486etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 487{
270 else 488 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
489
490 pthread_once (&doinit, etp_once_init);
491
492 want_poll_cb = want_poll;
493 done_poll_cb = done_poll;
494
271 return 0; 495 return 0;
272} 496}
273 497
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 498X_THREAD_PROC (etp_proc);
325 499
326static void start_thread (void) 500static void ecb_cold
501etp_start_thread (void)
327{ 502{
328 worker *wrk = calloc (1, sizeof (worker)); 503 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 504
330 /*TODO*/ 505 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 506 assert (("unable to allocate worker thread data", wrk));
332 507
333 X_LOCK (wrklock); 508 X_LOCK (wrklock);
334 509
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 510 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 511 {
337 wrk->prev = &wrk_first; 512 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 513 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 514 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 515 wrk_first.next = wrk;
344 free (wrk); 519 free (wrk);
345 520
346 X_UNLOCK (wrklock); 521 X_UNLOCK (wrklock);
347} 522}
348 523
524static void
349static void maybe_start_thread (void) 525etp_maybe_start_thread (void)
350{ 526{
351 if (eio_nthreads () >= wanted) 527 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 528 return;
353 529
354 /* todo: maybe use idle here, but might be less exact */ 530 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 531 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 532 return;
357 533
358 start_thread (); 534 etp_start_thread ();
359} 535}
360 536
361void eio_submit (eio_req *req) 537static void ecb_cold
362{ 538etp_end_thread (void)
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void)
375{ 539{
376 eio_req *req = calloc (1, sizeof (eio_req)); 540 eio_req *req = calloc (1, sizeof (eio_req));
377 541
378 req->type = EIO_QUIT; 542 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 543 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 544
381 X_LOCK (reqlock); 545 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 546 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 547 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 548 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 550 X_LOCK (wrklock);
387 --started; 551 --started;
388 X_UNLOCK (wrklock); 552 X_UNLOCK (wrklock);
389} 553}
390 554
391void eio_set_max_poll_time (double nseconds) 555static int
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 556etp_poll (void)
428{ 557{
429 int maxreqs = max_poll_reqs; 558 unsigned int maxreqs;
559 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 560 struct timeval tv_start, tv_now;
431 eio_req *req;
432 561
562 X_LOCK (reslock);
563 maxreqs = max_poll_reqs;
564 maxtime = max_poll_time;
565 X_UNLOCK (reslock);
566
433 if (max_poll_time) 567 if (maxtime)
434 gettimeofday (&tv_start, 0); 568 gettimeofday (&tv_start, 0);
435 569
436 for (;;) 570 for (;;)
437 { 571 {
572 ETP_REQ *req;
573
438 maybe_start_thread (); 574 etp_maybe_start_thread ();
439 575
440 X_LOCK (reslock); 576 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 577 req = reqq_shift (&res_queue);
442 578
443 if (req) 579 if (req)
451 X_UNLOCK (reslock); 587 X_UNLOCK (reslock);
452 588
453 if (!req) 589 if (!req)
454 return 0; 590 return 0;
455 591
592 X_LOCK (reqlock);
456 --nreqs; 593 --nreqs;
594 X_UNLOCK (reqlock);
457 595
458 if (req->type == EIO_GROUP && req->size) 596 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 597 {
460 req->int1 = 1; /* mark request as delayed */ 598 req->int1 = 1; /* mark request as delayed */
461 continue; 599 continue;
462 } 600 }
463 else 601 else
464 { 602 {
465 int res = eio_finish (req); 603 int res = ETP_FINISH (req);
466 if (res) 604 if (ecb_expect_false (res))
467 return res; 605 return res;
468 } 606 }
469 607
470 if (maxreqs && !--maxreqs) 608 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 609 break;
472 610
473 if (max_poll_time) 611 if (maxtime)
474 { 612 {
475 gettimeofday (&tv_now, 0); 613 gettimeofday (&tv_now, 0);
476 614
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 615 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 616 break;
479 } 617 }
480 } 618 }
481 619
482 errno = EAGAIN; 620 errno = EAGAIN;
483 return -1; 621 return -1;
484} 622}
485 623
624static void
625etp_cancel (ETP_REQ *req)
626{
627 req->cancelled = 1;
628
629 eio_grp_cancel (req);
630}
631
632static void
633etp_submit (ETP_REQ *req)
634{
635 req->pri -= ETP_PRI_MIN;
636
637 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
638 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
639
640 if (ecb_expect_false (req->type == EIO_GROUP))
641 {
642 /* I hope this is worth it :/ */
643 X_LOCK (reqlock);
644 ++nreqs;
645 X_UNLOCK (reqlock);
646
647 X_LOCK (reslock);
648
649 ++npending;
650
651 if (!reqq_push (&res_queue, req) && want_poll_cb)
652 want_poll_cb ();
653
654 X_UNLOCK (reslock);
655 }
656 else
657 {
658 X_LOCK (reqlock);
659 ++nreqs;
660 ++nready;
661 reqq_push (&req_queue, req);
662 X_COND_SIGNAL (reqwait);
663 X_UNLOCK (reqlock);
664
665 etp_maybe_start_thread ();
666 }
667}
668
669static void ecb_cold
670etp_set_max_poll_time (double nseconds)
671{
672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
673 max_poll_time = nseconds * EIO_TICKS;
674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
675}
676
677static void ecb_cold
678etp_set_max_poll_reqs (unsigned int maxreqs)
679{
680 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
681 max_poll_reqs = maxreqs;
682 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
683}
684
685static void ecb_cold
686etp_set_max_idle (unsigned int nthreads)
687{
688 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
689 max_idle = nthreads;
690 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
691}
692
693static void ecb_cold
694etp_set_idle_timeout (unsigned int seconds)
695{
696 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
697 idle_timeout = seconds;
698 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
699}
700
701static void ecb_cold
702etp_set_min_parallel (unsigned int nthreads)
703{
704 if (wanted < nthreads)
705 wanted = nthreads;
706}
707
708static void ecb_cold
709etp_set_max_parallel (unsigned int nthreads)
710{
711 if (wanted > nthreads)
712 wanted = nthreads;
713
714 while (started > wanted)
715 etp_end_thread ();
716}
717
718/*****************************************************************************/
719
720static void
721grp_try_feed (eio_req *grp)
722{
723 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
724 {
725 grp->flags &= ~EIO_FLAG_GROUPADD;
726
727 EIO_FEED (grp);
728
729 /* stop if no progress has been made */
730 if (!(grp->flags & EIO_FLAG_GROUPADD))
731 {
732 grp->feed = 0;
733 break;
734 }
735 }
736}
737
738static int
739grp_dec (eio_req *grp)
740{
741 --grp->size;
742
743 /* call feeder, if applicable */
744 grp_try_feed (grp);
745
746 /* finish, if done */
747 if (!grp->size && grp->int1)
748 return eio_finish (grp);
749 else
750 return 0;
751}
752
753static void
754eio_destroy (eio_req *req)
755{
756 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
757 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
758
759 EIO_DESTROY (req);
760}
761
762static int
763eio_finish (eio_req *req)
764{
765 int res = EIO_FINISH (req);
766
767 if (req->grp)
768 {
769 int res2;
770 eio_req *grp = req->grp;
771
772 /* unlink request */
773 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
774 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
775
776 if (grp->grp_first == req)
777 grp->grp_first = req->grp_next;
778
779 res2 = grp_dec (grp);
780
781 if (!res)
782 res = res2;
783 }
784
785 eio_destroy (req);
786
787 return res;
788}
789
790void
791eio_grp_cancel (eio_req *grp)
792{
793 for (grp = grp->grp_first; grp; grp = grp->grp_next)
794 eio_cancel (grp);
795}
796
797void
798eio_cancel (eio_req *req)
799{
800 etp_cancel (req);
801}
802
803void
804eio_submit (eio_req *req)
805{
806 etp_submit (req);
807}
808
809unsigned int
810eio_nreqs (void)
811{
812 return etp_nreqs ();
813}
814
815unsigned int
816eio_nready (void)
817{
818 return etp_nready ();
819}
820
821unsigned int
822eio_npending (void)
823{
824 return etp_npending ();
825}
826
827unsigned int ecb_cold
828eio_nthreads (void)
829{
830 return etp_nthreads ();
831}
832
833void ecb_cold
834eio_set_max_poll_time (double nseconds)
835{
836 etp_set_max_poll_time (nseconds);
837}
838
839void ecb_cold
840eio_set_max_poll_reqs (unsigned int maxreqs)
841{
842 etp_set_max_poll_reqs (maxreqs);
843}
844
845void ecb_cold
846eio_set_max_idle (unsigned int nthreads)
847{
848 etp_set_max_idle (nthreads);
849}
850
851void ecb_cold
852eio_set_idle_timeout (unsigned int seconds)
853{
854 etp_set_idle_timeout (seconds);
855}
856
857void ecb_cold
858eio_set_min_parallel (unsigned int nthreads)
859{
860 etp_set_min_parallel (nthreads);
861}
862
863void ecb_cold
864eio_set_max_parallel (unsigned int nthreads)
865{
866 etp_set_max_parallel (nthreads);
867}
868
869int eio_poll (void)
870{
871 return etp_poll ();
872}
873
486/*****************************************************************************/ 874/*****************************************************************************/
487/* work around various missing functions */ 875/* work around various missing functions */
488 876
489#if !HAVE_PREADWRITE 877#if !HAVE_PREADWRITE
878# undef pread
879# undef pwrite
490# define pread aio_pread 880# define pread eio__pread
491# define pwrite aio_pwrite 881# define pwrite eio__pwrite
492 882
493/* 883static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 884eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 885{
502 ssize_t res; 886 ssize_t res;
503 off_t ooffset; 887 off_t ooffset;
504 888
505 X_LOCK (preadwritelock); 889 X_LOCK (preadwritelock);
510 X_UNLOCK (preadwritelock); 894 X_UNLOCK (preadwritelock);
511 895
512 return res; 896 return res;
513} 897}
514 898
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 899static ssize_t
900eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 901{
517 ssize_t res; 902 ssize_t res;
518 off_t ooffset; 903 off_t ooffset;
519 904
520 X_LOCK (preadwritelock); 905 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 906 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 907 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 908 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 909 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 910 X_UNLOCK (preadwritelock);
526 911
527 return res; 912 return res;
528} 913}
529#endif 914#endif
530 915
531#ifndef HAVE_FUTIMES 916#ifndef HAVE_UTIMES
532 917
918# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 919# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 920
921static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 922eio__utimes (const char *filename, const struct timeval times[2])
537{ 923{
538 if (times) 924 if (times)
539 { 925 {
540 struct utimbuf buf; 926 struct utimbuf buf;
541 927
546 } 932 }
547 else 933 else
548 return utime (filename, 0); 934 return utime (filename, 0);
549} 935}
550 936
937#endif
938
939#ifndef HAVE_FUTIMES
940
941# undef futimes
942# define futimes(fd,times) eio__futimes (fd, times)
943
944static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 945eio__futimes (int fd, const struct timeval tv[2])
552{ 946{
553 errno = ENOSYS; 947 errno = ENOSYS;
554 return -1; 948 return -1;
555} 949}
556 950
557#endif 951#endif
558 952
559#if !HAVE_FDATASYNC 953#if !HAVE_FDATASYNC
954# undef fdatasync
560# define fdatasync fsync 955# define fdatasync(fd) fsync (fd)
561#endif 956#endif
957
958/* sync_file_range always needs emulation */
959static int
960eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
961{
962#if HAVE_SYNC_FILE_RANGE
963 int res;
964
965 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
966 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
967 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
968 {
969 flags = 0
970 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
971 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
972 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
973 }
974
975 res = sync_file_range (fd, offset, nbytes, flags);
976
977 if (!res || errno != ENOSYS)
978 return res;
979#endif
980
981 /* even though we could play tricks with the flags, it's better to always
982 * call fdatasync, as that matches the expectation of its users best */
983 return fdatasync (fd);
984}
985
986static int
987eio__fallocate (int fd, int mode, off_t offset, size_t len)
988{
989#if HAVE_FALLOCATE
990 return fallocate (fd, mode, offset, len);
991#else
992 errno = ENOSYS;
993 return -1;
994#endif
995}
562 996
563#if !HAVE_READAHEAD 997#if !HAVE_READAHEAD
998# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 999# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 1000
1001static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1002eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 1003{
568 size_t todo = count; 1004 size_t todo = count;
569 dBUF; 1005 dBUF;
570 1006
571 while (todo > 0) 1007 while (todo > 0)
581 return count; 1017 return count;
582} 1018}
583 1019
584#endif 1020#endif
585 1021
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif
615
616/* sendfile always needs emulation */ 1022/* sendfile always needs emulation */
1023static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1024eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 1025{
1026 ssize_t written = 0;
619 ssize_t res; 1027 ssize_t res;
620 1028
621 if (!count) 1029 if (!count)
622 return 0; 1030 return 0;
623 1031
1032 for (;;)
1033 {
1034#ifdef __APPLE__
1035# undef HAVE_SENDFILE /* broken, as everything on os x */
1036#endif
624#if HAVE_SENDFILE 1037#if HAVE_SENDFILE
625# if __linux 1038# if __linux
1039 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1040 res = sendfile (ofd, ifd, &soffset, count);
627 1041
628# elif __freebsd 1042# elif __FreeBSD__
629 /* 1043 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 1044 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1045 * wasted on making it similar to other I/O functions.
632 */ 1046 */
633 {
634 off_t sbytes; 1047 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1048 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1049
637 if (res < 0 && sbytes) 1050 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1051 /* freebsd' sendfile will return 0 on success */
1052 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1053 /* not on e.g. EIO or EPIPE - sounds broken */
1054 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1055 res = sbytes;
640 } 1056 #endif
1057
1058 /* according to source inspection, this is correct, and useful behaviour */
1059 if (sbytes)
1060 res = sbytes;
1061
1062# elif defined (__APPLE__)
1063 off_t sbytes = count;
1064 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1065
1066 /* according to the manpage, sbytes is always valid */
1067 if (sbytes)
1068 res = sbytes;
641 1069
642# elif __hpux 1070# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1071 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1072
645# elif __solaris 1073# elif __solaris
646 {
647 struct sendfilevec vec; 1074 struct sendfilevec vec;
648 size_t sbytes; 1075 size_t sbytes;
649 1076
650 vec.sfv_fd = ifd; 1077 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1078 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1079 vec.sfv_off = offset;
653 vec.sfv_len = count; 1080 vec.sfv_len = count;
654 1081
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1082 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1083
657 if (res < 0 && sbytes) 1084 if (res < 0 && sbytes)
658 res = sbytes; 1085 res = sbytes;
659 }
660 1086
661# endif 1087# endif
1088
1089#elif defined (_WIN32) && 0
1090 /* does not work, just for documentation of what would need to be done */
1091 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1092 /* libeio guarantees that the file offset does not change, and windows */
1093 /* has no way to get an independent handle to the same file description */
1094 HANDLE h = TO_SOCKET (ifd);
1095 SetFilePointer (h, offset, 0, FILE_BEGIN);
1096 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1097
662#else 1098#else
663 res = -1; 1099 res = -1;
664 errno = ENOSYS; 1100 errno = ENOSYS;
665#endif 1101#endif
666 1102
1103 /* we assume sendfile can copy at least 128mb in one go */
1104 if (res <= 128 * 1024 * 1024)
1105 {
1106 if (res > 0)
1107 written += res;
1108
1109 if (written)
1110 return written;
1111
1112 break;
1113 }
1114 else
1115 {
1116 /* if we requested more, then probably the kernel was lazy */
1117 written += res;
1118 offset += res;
1119 count -= res;
1120
1121 if (!count)
1122 return written;
1123 }
1124 }
1125
667 if (res < 0 1126 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1127 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1128 /* BSDs */
1129#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1130 || errno == ENOTSUP
1131#endif
1132 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1133#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1134 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1135#endif
672 ) 1136 )
673 ) 1137 )
704 } 1168 }
705 1169
706 return res; 1170 return res;
707} 1171}
708 1172
1173#ifdef PAGESIZE
1174# define eio_pagesize() PAGESIZE
1175#else
1176static intptr_t
1177eio_pagesize (void)
1178{
1179 static intptr_t page;
1180
1181 if (!page)
1182 page = sysconf (_SC_PAGESIZE);
1183
1184 return page;
1185}
1186#endif
1187
1188static void
1189eio_page_align (void **addr, size_t *length)
1190{
1191 intptr_t mask = eio_pagesize () - 1;
1192
1193 /* round down addr */
1194 intptr_t adj = mask & (intptr_t)*addr;
1195
1196 *addr = (void *)((intptr_t)*addr - adj);
1197 *length += adj;
1198
1199 /* round up length */
1200 *length = (*length + mask) & ~mask;
1201}
1202
1203#if !_POSIX_MEMLOCK
1204# define eio__mlockall(a) eio_nosyscall()
1205#else
1206
1207static int
1208eio__mlockall (int flags)
1209{
1210 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1211 extern int mallopt (int, int);
1212 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1213 #endif
1214
1215 if (EIO_MCL_CURRENT != MCL_CURRENT
1216 || EIO_MCL_FUTURE != MCL_FUTURE)
1217 {
1218 flags = 0
1219 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1220 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1221 }
1222
1223 return mlockall (flags);
1224}
1225#endif
1226
1227#if !_POSIX_MEMLOCK_RANGE
1228# define eio__mlock(a,b) EIO_ENOSYS ()
1229#else
1230
1231static int
1232eio__mlock (void *addr, size_t length)
1233{
1234 eio_page_align (&addr, &length);
1235
1236 return mlock (addr, length);
1237}
1238
1239#endif
1240
1241#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1242# define eio__msync(a,b,c) EIO_ENOSYS ()
1243#else
1244
1245static int
1246eio__msync (void *mem, size_t len, int flags)
1247{
1248 eio_page_align (&mem, &len);
1249
1250 if (EIO_MS_ASYNC != MS_SYNC
1251 || EIO_MS_INVALIDATE != MS_INVALIDATE
1252 || EIO_MS_SYNC != MS_SYNC)
1253 {
1254 flags = 0
1255 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1256 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1257 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1258 }
1259
1260 return msync (mem, len, flags);
1261}
1262
1263#endif
1264
1265static int
1266eio__mtouch (eio_req *req)
1267{
1268 void *mem = req->ptr2;
1269 size_t len = req->size;
1270 int flags = req->int1;
1271
1272 eio_page_align (&mem, &len);
1273
1274 {
1275 intptr_t addr = (intptr_t)mem;
1276 intptr_t end = addr + len;
1277 intptr_t page = eio_pagesize ();
1278
1279 if (addr < end)
1280 if (flags & EIO_MT_MODIFY) /* modify */
1281 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1282 else
1283 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1284 }
1285
1286 return 0;
1287}
1288
1289/*****************************************************************************/
1290/* requests implemented outside eio_execute, because they are so large */
1291
1292static void
1293eio__realpath (eio_req *req, etp_worker *self)
1294{
1295 char *rel = req->ptr1;
1296 char *res;
1297 char *tmp1, *tmp2;
1298#if SYMLOOP_MAX > 32
1299 int symlinks = SYMLOOP_MAX;
1300#else
1301 int symlinks = 32;
1302#endif
1303
1304 req->result = -1;
1305
1306 errno = EINVAL;
1307 if (!rel)
1308 return;
1309
1310 errno = ENOENT;
1311 if (!*rel)
1312 return;
1313
1314 if (!req->ptr2)
1315 {
1316 X_LOCK (wrklock);
1317 req->flags |= EIO_FLAG_PTR2_FREE;
1318 X_UNLOCK (wrklock);
1319 req->ptr2 = malloc (PATH_MAX * 3);
1320
1321 errno = ENOMEM;
1322 if (!req->ptr2)
1323 return;
1324 }
1325
1326 res = req->ptr2;
1327 tmp1 = res + PATH_MAX;
1328 tmp2 = tmp1 + PATH_MAX;
1329
1330#if 0 /* disabled, the musl way to do things is just too racy */
1331#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1332 /* on linux we may be able to ask the kernel */
1333 {
1334 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1335
1336 if (fd >= 0)
1337 {
1338 sprintf (tmp1, "/proc/self/fd/%d", fd);
1339 req->result = readlink (tmp1, res, PATH_MAX);
1340 close (fd);
1341
1342 /* here we should probably stat the open file and the disk file, to make sure they still match */
1343
1344 if (req->result > 0)
1345 goto done;
1346 }
1347 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1348 return;
1349 }
1350#endif
1351#endif
1352
1353 if (*rel != '/')
1354 {
1355 if (!getcwd (res, PATH_MAX))
1356 return;
1357
1358 if (res [1]) /* only use if not / */
1359 res += strlen (res);
1360 }
1361
1362 while (*rel)
1363 {
1364 ssize_t len, linklen;
1365 char *beg = rel;
1366
1367 while (*rel && *rel != '/')
1368 ++rel;
1369
1370 len = rel - beg;
1371
1372 if (!len) /* skip slashes */
1373 {
1374 ++rel;
1375 continue;
1376 }
1377
1378 if (beg [0] == '.')
1379 {
1380 if (len == 1)
1381 continue; /* . - nop */
1382
1383 if (beg [1] == '.' && len == 2)
1384 {
1385 /* .. - back up one component, if possible */
1386
1387 while (res != req->ptr2)
1388 if (*--res == '/')
1389 break;
1390
1391 continue;
1392 }
1393 }
1394
1395 errno = ENAMETOOLONG;
1396 if (res + 1 + len + 1 >= tmp1)
1397 return;
1398
1399 /* copy one component */
1400 *res = '/';
1401 memcpy (res + 1, beg, len);
1402
1403 /* zero-terminate, for readlink */
1404 res [len + 1] = 0;
1405
1406 /* now check if it's a symlink */
1407 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1408
1409 if (linklen < 0)
1410 {
1411 if (errno != EINVAL)
1412 return;
1413
1414 /* it's a normal directory. hopefully */
1415 res += len + 1;
1416 }
1417 else
1418 {
1419 /* yay, it was a symlink - build new path in tmp2 */
1420 int rellen = strlen (rel);
1421
1422 errno = ENAMETOOLONG;
1423 if (linklen + 1 + rellen >= PATH_MAX)
1424 return;
1425
1426 errno = ELOOP;
1427 if (!--symlinks)
1428 return;
1429
1430 if (*tmp1 == '/')
1431 res = req->ptr2; /* symlink resolves to an absolute path */
1432
1433 /* we need to be careful, as rel might point into tmp2 already */
1434 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1435 tmp2 [linklen] = '/';
1436 memcpy (tmp2, tmp1, linklen);
1437
1438 rel = tmp2;
1439 }
1440 }
1441
1442 /* special case for the lone root path */
1443 if (res == req->ptr2)
1444 *res++ = '/';
1445
1446 req->result = res - (char *)req->ptr2;
1447
1448done:
1449 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1450}
1451
1452static signed char
1453eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1454{
1455 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1456 : a->inode < b->inode ? -1
1457 : a->inode > b->inode ? 1
1458 : 0;
1459}
1460
1461#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1462
1463#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1464#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1465
1466static void
1467eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1468{
1469 unsigned char bits [9 + sizeof (ino_t) * 8];
1470 unsigned char *bit = bits;
1471
1472 assert (CHAR_BIT == 8);
1473 assert (sizeof (eio_dirent) * 8 < 256);
1474 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1475 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1476
1477 if (size <= EIO_SORT_FAST)
1478 return;
1479
1480 /* first prepare an array of bits to test in our radix sort */
1481 /* try to take endianness into account, as well as differences in ino_t sizes */
1482 /* inode_bits must contain all inodes ORed together */
1483 /* which is used to skip bits that are 0 everywhere, which is very common */
1484 {
1485 ino_t endianness;
1486 int i, j;
1487
1488 /* we store the byte offset of byte n into byte n of "endianness" */
1489 for (i = 0; i < sizeof (ino_t); ++i)
1490 ((unsigned char *)&endianness)[i] = i;
1491
1492 *bit++ = 0;
1493
1494 for (i = 0; i < sizeof (ino_t); ++i)
1495 {
1496 /* shifting off the byte offsets out of "endianness" */
1497 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1498 endianness >>= 8;
1499
1500 for (j = 0; j < 8; ++j)
1501 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1502 *bit++ = offs + j;
1503 }
1504
1505 for (j = 0; j < 8; ++j)
1506 if (score_bits & (1 << j))
1507 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1508 }
1509
1510 /* now actually do the sorting (a variant of MSD radix sort) */
1511 {
1512 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1513 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1514 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1515 int stk_idx = 0;
1516
1517 base_stk [stk_idx] = dents;
1518 end_stk [stk_idx] = dents + size;
1519 bit_stk [stk_idx] = bit - 1;
1520
1521 do
1522 {
1523 base = base_stk [stk_idx];
1524 end = end_stk [stk_idx];
1525 bit = bit_stk [stk_idx];
1526
1527 for (;;)
1528 {
1529 unsigned char O = *bit >> 3;
1530 unsigned char M = 1 << (*bit & 7);
1531
1532 eio_dirent *a = base;
1533 eio_dirent *b = end;
1534
1535 if (b - a < EIO_SORT_CUTOFF)
1536 break;
1537
1538 /* now bit-partition the array on the bit */
1539 /* this ugly asymmetric loop seems to perform much better than typical */
1540 /* partition algos found in the literature */
1541 do
1542 if (!(((unsigned char *)a)[O] & M))
1543 ++a;
1544 else if (!(((unsigned char *)--b)[O] & M))
1545 {
1546 eio_dirent tmp = *a; *a = *b; *b = tmp;
1547 ++a;
1548 }
1549 while (b > a);
1550
1551 /* next bit, or stop, if no bits left in this path */
1552 if (!*--bit)
1553 break;
1554
1555 base_stk [stk_idx] = a;
1556 end_stk [stk_idx] = end;
1557 bit_stk [stk_idx] = bit;
1558 ++stk_idx;
1559
1560 end = a;
1561 }
1562 }
1563 while (stk_idx--);
1564 }
1565}
1566
1567static void
1568eio_dent_insertion_sort (eio_dirent *dents, int size)
1569{
1570 /* first move the smallest element to the front, to act as a sentinel */
1571 {
1572 int i;
1573 eio_dirent *min = dents;
1574
1575 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1576 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1577 if (EIO_DENT_CMP (dents [i], <, *min))
1578 min = &dents [i];
1579
1580 /* swap elements 0 and j (minimum) */
1581 {
1582 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1583 }
1584 }
1585
1586 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1587 {
1588 eio_dirent *i, *j;
1589
1590 for (i = dents + 1; i < dents + size; ++i)
1591 {
1592 eio_dirent value = *i;
1593
1594 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1595 j [1] = j [0];
1596
1597 j [1] = value;
1598 }
1599 }
1600}
1601
1602static void
1603eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1604{
1605 if (size <= 1)
1606 return; /* our insertion sort relies on size > 0 */
1607
1608 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1609 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1610 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1611
1612 /* use an insertion sort at the end, or for small arrays, */
1613 /* as insertion sort is more efficient for small partitions */
1614 eio_dent_insertion_sort (dents, size);
1615}
1616
709/* read a full directory */ 1617/* read a full directory */
1618static void
710static void scandir_ (eio_req *req, worker *self) 1619eio__scandir (eio_req *req, etp_worker *self)
711{ 1620{
712 DIR *dirp; 1621 DIR *dirp;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp; 1622 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1623 char *name, *names;
720 int memlen = 4096; 1624 int namesalloc = 4096;
721 int memofs = 0; 1625 int namesoffs = 0;
1626 int flags = req->int1;
1627 eio_dirent *dents = 0;
1628 int dentalloc = 128;
722 int res = 0; 1629 int dentoffs = 0;
1630 ino_t inode_bits = 0;
1631
1632 req->result = -1;
1633
1634 if (!(flags & EIO_READDIR_DENTS))
1635 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1636
724 X_LOCK (wrklock); 1637 X_LOCK (wrklock);
1638 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1639 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1640
727 req->flags |= EIO_FLAG_PTR2_FREE; 1641 if (req->flags & EIO_FLAG_PTR1_FREE)
1642 free (req->ptr1);
1643
1644 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1645 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1646 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1647 X_UNLOCK (wrklock);
730 1648
731 if (dirp && u && names) 1649 if (dirp && names && (!flags || dents))
732 for (;;) 1650 for (;;)
733 { 1651 {
734 errno = 0; 1652 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1653 entp = readdir (dirp);
736 1654
737 if (!entp) 1655 if (!entp)
1656 {
1657 if (errno)
1658 break;
1659
1660 /* sort etc. */
1661 req->int1 = flags;
1662 req->result = dentoffs;
1663
1664 if (flags & EIO_READDIR_STAT_ORDER)
1665 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1666 else if (flags & EIO_READDIR_DIRS_FIRST)
1667 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1668 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1669 else
1670 {
1671 /* in this case, all is known, and we just put dirs first and sort them */
1672 eio_dirent *oth = dents + dentoffs;
1673 eio_dirent *dir = dents;
1674
1675 /* now partition dirs to the front, and non-dirs to the back */
1676 /* by walking from both sides and swapping if necessary */
1677 while (oth > dir)
1678 {
1679 if (dir->type == EIO_DT_DIR)
1680 ++dir;
1681 else if ((--oth)->type == EIO_DT_DIR)
1682 {
1683 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1684
1685 ++dir;
1686 }
1687 }
1688
1689 /* now sort the dirs only (dirs all have the same score) */
1690 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1691 }
1692
738 break; 1693 break;
1694 }
739 1695
1696 /* now add the entry to our list(s) */
740 name = entp->d_name; 1697 name = entp->d_name;
741 1698
1699 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1700 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1701 {
744 int len = strlen (name) + 1; 1702 int len = D_NAMLEN (entp) + 1;
745 1703
746 res++; 1704 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1705 {
750 memlen *= 2; 1706 namesalloc *= 2;
751 X_LOCK (wrklock); 1707 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1708 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1709 X_UNLOCK (wrklock);
754 1710
755 if (!names) 1711 if (!names)
756 break; 1712 break;
757 } 1713 }
758 1714
759 memcpy (names + memofs, name, len); 1715 memcpy (names + namesoffs, name, len);
1716
1717 if (dents)
1718 {
1719 struct eio_dirent *ent;
1720
1721 if (ecb_expect_false (dentoffs == dentalloc))
1722 {
1723 dentalloc *= 2;
1724 X_LOCK (wrklock);
1725 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1726 X_UNLOCK (wrklock);
1727
1728 if (!dents)
1729 break;
1730 }
1731
1732 ent = dents + dentoffs;
1733
1734 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1735 ent->namelen = len - 1;
1736 ent->inode = D_INO (entp);
1737
1738 inode_bits |= ent->inode;
1739
1740 switch (D_TYPE (entp))
1741 {
1742 default:
1743 ent->type = EIO_DT_UNKNOWN;
1744 flags |= EIO_READDIR_FOUND_UNKNOWN;
1745 break;
1746
1747 #ifdef DT_FIFO
1748 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1749 #endif
1750 #ifdef DT_CHR
1751 case DT_CHR: ent->type = EIO_DT_CHR; break;
1752 #endif
1753 #ifdef DT_MPC
1754 case DT_MPC: ent->type = EIO_DT_MPC; break;
1755 #endif
1756 #ifdef DT_DIR
1757 case DT_DIR: ent->type = EIO_DT_DIR; break;
1758 #endif
1759 #ifdef DT_NAM
1760 case DT_NAM: ent->type = EIO_DT_NAM; break;
1761 #endif
1762 #ifdef DT_BLK
1763 case DT_BLK: ent->type = EIO_DT_BLK; break;
1764 #endif
1765 #ifdef DT_MPB
1766 case DT_MPB: ent->type = EIO_DT_MPB; break;
1767 #endif
1768 #ifdef DT_REG
1769 case DT_REG: ent->type = EIO_DT_REG; break;
1770 #endif
1771 #ifdef DT_NWK
1772 case DT_NWK: ent->type = EIO_DT_NWK; break;
1773 #endif
1774 #ifdef DT_CMP
1775 case DT_CMP: ent->type = EIO_DT_CMP; break;
1776 #endif
1777 #ifdef DT_LNK
1778 case DT_LNK: ent->type = EIO_DT_LNK; break;
1779 #endif
1780 #ifdef DT_SOCK
1781 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1782 #endif
1783 #ifdef DT_DOOR
1784 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1785 #endif
1786 #ifdef DT_WHT
1787 case DT_WHT: ent->type = EIO_DT_WHT; break;
1788 #endif
1789 }
1790
1791 ent->score = 7;
1792
1793 if (flags & EIO_READDIR_DIRS_FIRST)
1794 {
1795 if (ent->type == EIO_DT_UNKNOWN)
1796 {
1797 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1798 ent->score = 1;
1799 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1800 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1801 }
1802 else if (ent->type == EIO_DT_DIR)
1803 ent->score = 0;
1804 }
1805 }
1806
760 memofs += len; 1807 namesoffs += len;
1808 ++dentoffs;
1809 }
1810
1811 if (EIO_CANCELLED (req))
1812 {
1813 errno = ECANCELED;
1814 break;
761 } 1815 }
762 } 1816 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1817}
769 1818
770/*****************************************************************************/ 1819/*****************************************************************************/
771 1820
772#define ALLOC(len) \ 1821#define ALLOC(len) \
782 req->result = -1; \ 1831 req->result = -1; \
783 break; \ 1832 break; \
784 } \ 1833 } \
785 } 1834 }
786 1835
787X_THREAD_PROC (eio_proc) 1836X_THREAD_PROC (etp_proc)
788{ 1837{
789 eio_req *req; 1838 ETP_REQ *req;
790 struct timespec ts; 1839 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1840 etp_worker *self = (etp_worker *)thr_arg;
1841 int timeout;
792 1842
793 /* try to distribute timeouts somewhat randomly */ 1843 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1844 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1845
796 for (;;) 1846 for (;;)
797 { 1847 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1848 ts.tv_sec = 0;
799 1849
800 X_LOCK (reqlock); 1850 X_LOCK (reqlock);
801 1851
802 for (;;) 1852 for (;;)
803 { 1853 {
804 self->req = req = reqq_shift (&req_queue); 1854 self->req = req = reqq_shift (&req_queue);
805 1855
806 if (req) 1856 if (req)
807 break; 1857 break;
808 1858
1859 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1860 {
1861 X_UNLOCK (reqlock);
1862 X_LOCK (wrklock);
1863 --started;
1864 X_UNLOCK (wrklock);
1865 goto quit;
1866 }
1867
809 ++idle; 1868 ++idle;
810 1869
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1870 if (idle <= max_idle)
1871 /* we are allowed to idle, so do so without any timeout */
1872 X_COND_WAIT (reqwait, reqlock);
1873 else
812 { 1874 {
813 if (idle > max_idle) 1875 /* initialise timeout once */
814 { 1876 if (!ts.tv_sec)
815 --idle; 1877 ts.tv_sec = time (0) + idle_timeout;
816 X_UNLOCK (reqlock);
817 X_LOCK (wrklock);
818 --started;
819 X_UNLOCK (wrklock);
820 goto quit;
821 }
822 1878
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1879 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1880 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 1881 }
827 1882
828 --idle; 1883 --idle;
829 } 1884 }
830 1885
831 --nready; 1886 --nready;
832 1887
833 X_UNLOCK (reqlock); 1888 X_UNLOCK (reqlock);
834 1889
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 1890 if (req->type < 0)
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit; 1891 goto quit;
930 1892
931 default: 1893 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1894
938 X_LOCK (reslock); 1895 X_LOCK (reslock);
939 1896
940 ++npending; 1897 ++npending;
941 1898
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1899 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1900 want_poll_cb ();
944 1901
945 self->req = 0; 1902 self->req = 0;
946 worker_clear (self); 1903 etp_worker_clear (self);
947 1904
948 X_UNLOCK (reslock); 1905 X_UNLOCK (reslock);
949 } 1906 }
950 1907
951quit: 1908quit:
952 X_LOCK (wrklock); 1909 X_LOCK (wrklock);
953 worker_free (self); 1910 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1911 X_UNLOCK (wrklock);
955 1912
956 return 0; 1913 return 0;
957} 1914}
958 1915
959/*****************************************************************************/ 1916/*****************************************************************************/
960 1917
961static void eio_atfork_prepare (void) 1918int ecb_cold
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1919eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1920{
1019 want_poll_cb = want_poll; 1921 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 1922}
1035 1923
1924ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1925eio_api_destroy (eio_req *req)
1037{ 1926{
1038 free (req); 1927 free (req);
1039} 1928}
1040 1929
1041#define REQ(rtype) \ 1930#define REQ(rtype) \
1043 \ 1932 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1933 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1934 if (!req) \
1046 return 0; \ 1935 return 0; \
1047 \ 1936 \
1048 req->type = EIO_ ## rtype; \ 1937 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1938 req->pri = pri; \
1050 req->finish = cb; \ 1939 req->finish = cb; \
1940 req->data = data; \
1051 req->destroy = eio_api_destroy; 1941 req->destroy = eio_api_destroy;
1052 1942
1053#define SEND eio_submit (req); return req 1943#define SEND eio_submit (req); return req
1054 1944
1055#define PATH (void)0 1945#define PATH \
1946 req->flags |= EIO_FLAG_PTR1_FREE; \
1947 req->ptr1 = strdup (path); \
1948 if (!req->ptr1) \
1949 { \
1950 eio_api_destroy (req); \
1951 return 0; \
1952 }
1056 1953
1954static void
1955eio_execute (etp_worker *self, eio_req *req)
1956{
1957 if (ecb_expect_false (EIO_CANCELLED (req)))
1958 {
1959 req->result = -1;
1960 req->errorno = ECANCELED;
1961 return;
1962 }
1963
1964 switch (req->type)
1965 {
1966 case EIO_READ: ALLOC (req->size);
1967 req->result = req->offs >= 0
1968 ? pread (req->int1, req->ptr2, req->size, req->offs)
1969 : read (req->int1, req->ptr2, req->size); break;
1970 case EIO_WRITE: req->result = req->offs >= 0
1971 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1972 : write (req->int1, req->ptr2, req->size); break;
1973
1974 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1975 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1976
1977 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1979 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1980 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1981 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1982 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1983
1984 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1987 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1988
1989 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1990 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1991 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1992 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1993 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1994 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1995
1996 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1997 case EIO_CLOSE: req->result = close (req->int1); break;
1998 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1999 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2000 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2001 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2002 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2003 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2004 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2005 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2006
2007 case EIO_REALPATH: eio__realpath (req, self); break;
2008
2009 case EIO_READLINK: ALLOC (PATH_MAX);
2010 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2011
2012 case EIO_SYNC: req->result = 0; sync (); break;
2013 case EIO_FSYNC: req->result = fsync (req->int1); break;
2014 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2015 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2016 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2017 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2018 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2019 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2020 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2021
2022 case EIO_READDIR: eio__scandir (req, self); break;
2023
2024 case EIO_BUSY:
2025#ifdef _WIN32
2026 Sleep (req->nv1 * 1e3);
2027#else
2028 {
2029 struct timeval tv;
2030
2031 tv.tv_sec = req->nv1;
2032 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2033
2034 req->result = select (0, 0, 0, 0, &tv);
2035 }
2036#endif
2037 break;
2038
2039 case EIO_UTIME:
2040 case EIO_FUTIME:
2041 {
2042 struct timeval tv[2];
2043 struct timeval *times;
2044
2045 if (req->nv1 != -1. || req->nv2 != -1.)
2046 {
2047 tv[0].tv_sec = req->nv1;
2048 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2049 tv[1].tv_sec = req->nv2;
2050 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2051
2052 times = tv;
2053 }
2054 else
2055 times = 0;
2056
2057 req->result = req->type == EIO_FUTIME
2058 ? futimes (req->int1, times)
2059 : utimes (req->ptr1, times);
2060 }
2061 break;
2062
2063 case EIO_GROUP:
2064 abort (); /* handled in eio_request */
2065
2066 case EIO_NOP:
2067 req->result = 0;
2068 break;
2069
2070 case EIO_CUSTOM:
2071 req->feed (req);
2072 break;
2073
2074 default:
2075 errno = ENOSYS;
2076 req->result = -1;
2077 break;
2078 }
2079
2080 req->errorno = errno;
2081}
2082
2083#ifndef EIO_NO_WRAPPERS
2084
2085eio_req *eio_nop (int pri, eio_cb cb, void *data)
2086{
2087 REQ (EIO_NOP); SEND;
2088}
2089
2090eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2091{
2092 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2093}
2094
1057eio_req *eio_fsync (int fd, eio_cb cb) 2095eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2096{
2097 REQ (EIO_SYNC); SEND;
2098}
2099
2100eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2101{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2102 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2103}
1061 2104
2105eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2106{
2107 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2108}
2109
2110eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2111{
2112 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2113}
2114
2115eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2116{
2117 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2118}
2119
2120eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2121{
2122 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2123}
2124
2125eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2126{
2127 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2128}
2129
2130eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2131{
2132 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2133}
2134
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2135eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2136{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2137 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2138}
1066 2139
2140eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2141{
2142 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2143}
2144
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2145eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2146{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2147 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2148}
1071 2149
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2150eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2151{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2152 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2153}
1076 2154
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2155eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2156{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2157 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2158}
1081 2159
1082eio_req *eio_fstat (int fd, eio_cb cb) 2160eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2161{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2162 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2163}
1086 2164
2165eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2166{
2167 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2168}
2169
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2170eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2171{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2172 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2173}
1091 2174
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2175eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2176{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2177 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2178}
1096 2179
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2180eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2181{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2182 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2183}
1101 2184
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2185eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2186{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2187 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2188}
1106 2189
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2190eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2191{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2192 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2193}
1111 2194
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2195eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2196{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2197 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2198}
1116 2199
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2200eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2201{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2202 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2203}
1121 2204
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2205eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2206{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2207 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2208}
1141 2209
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2210eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2211{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2212 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2213}
1146 2214
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2215eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2216{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2217 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2218}
1151 2219
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2220eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2221{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2222 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2223}
1156 2224
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2225eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2226{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2227 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2228}
1161 2229
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2230static eio_req *
2231eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2232{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2233 REQ (type); PATH; SEND;
1170} 2234}
1171 2235
2236eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2237{
2238 return eio__1path (EIO_READLINK, path, pri, cb, data);
2239}
2240
2241eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2242{
2243 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2244}
2245
2246eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2247{
2248 return eio__1path (EIO_STAT, path, pri, cb, data);
2249}
2250
2251eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2252{
2253 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2254}
2255
2256eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2257{
2258 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2259}
2260
2261eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2262{
2263 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2264}
2265
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2266eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2267{
1174 REQ (READDIR); PATH; SEND; 2268 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2269}
1176 2270
2271eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2272{
2273 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2274}
2275
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2276eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2277{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2278 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2279}
1181 2280
1182eio_req *eio_busy (double delay, eio_cb cb) 2281static eio_req *
2282eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2283{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2284 REQ (type); PATH;
1185}
1186 2285
2286 req->flags |= EIO_FLAG_PTR2_FREE;
2287 req->ptr2 = strdup (new_path);
2288 if (!req->ptr2)
2289 {
2290 eio_api_destroy (req);
2291 return 0;
2292 }
2293
2294 SEND;
2295}
2296
2297eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2298{
2299 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2300}
2301
2302eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2303{
2304 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2305}
2306
2307eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2308{
2309 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2310}
2311
2312eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2313{
2314 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2315}
2316
2317#endif
2318
1187eio_req *eio_nop (eio_cb cb) 2319eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2320{
2321 const int pri = EIO_PRI_MAX;
2322
1189 REQ (NOP); SEND; 2323 REQ (EIO_GROUP); SEND;
1190} 2324}
1191 2325
1192#undef REQ 2326#undef REQ
1193#undef PATH 2327#undef PATH
1194#undef SEND 2328#undef SEND
1195 2329
1196#if 0 2330/*****************************************************************************/
2331/* grp functions */
1197 2332
1198void 2333void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2334eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2335{
1573 grp->int2 = limit; 2336 grp->int2 = limit;
1574 grp->feed = feed; 2337 grp->feed = feed;
1575 2338
1576 grp_try_feed (grp); 2339 grp_try_feed (grp);
1577} 2340}
1578 2341
2342void
1579void eio_grp_limit (eio_req *grp, int limit) 2343eio_grp_limit (eio_req *grp, int limit)
1580{ 2344{
1581 grp->int2 = limit; 2345 grp->int2 = limit;
1582 2346
1583 grp_try_feed (grp); 2347 grp_try_feed (grp);
1584} 2348}
1585 2349
2350void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2351eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2352{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2353 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2354
2355 grp->flags |= EIO_FLAG_GROUPADD;
1589 2356
1590 ++grp->size; 2357 ++grp->size;
1591 req->grp = grp; 2358 req->grp = grp;
1592 2359
1593 req->grp_prev = 0; 2360 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2364 grp->grp_first->grp_prev = req;
1598 2365
1599 grp->grp_first = req; 2366 grp->grp_first = req;
1600} 2367}
1601 2368
2369/*****************************************************************************/
2370/* misc garbage */
1602 2371
2372ssize_t
2373eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2374{
2375 etp_worker wrk;
2376 ssize_t ret;
2377
2378 wrk.dbuf = 0;
2379
2380 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2381
2382 if (wrk.dbuf)
2383 free (wrk.dbuf);
2384
2385 return ret;
2386}
2387

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines