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

Comparing libeio/eio.c (file contents):
Revision 1.12 by root, Tue May 13 18:54:52 2008 UTC vs.
Revision 1.89 by root, Thu Jul 14 22:36:17 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/* 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
73static void eio_destroy (eio_req *req);
74
15#ifndef EIO_FINISH 75#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 76# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 77#endif
18 78
19#ifndef EIO_DESTROY 79#ifndef EIO_DESTROY
… …
22 82
23#ifndef EIO_FEED 83#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 84# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 85#endif
26 86
87#ifndef EIO_FD_TO_WIN32_HANDLE
88# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
89#endif
90#ifndef EIO_WIN32_HANDLE_TO_FD
91# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
92#endif
93
94#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
95
96#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97
27#ifdef _WIN32 98#ifdef _WIN32
28 99
29 /*doh*/ 100 #define PAGESIZE 4096 /* GetSystemInfo? */
101
102 #ifdef EIO_STRUCT_STATI64
103 #define stat(path,buf) _stati64 (path,buf)
104 #define fstat(fd,buf) _fstati64 (path,buf)
105 #endif
106 #define lstat(path,buf) stat (path,buf)
107 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108 #define mkdir(path,mode) _mkdir (path)
109 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110
111 #define chown(path,uid,gid) EIO_ENOSYS ()
112 #define fchown(fd,uid,gid) EIO_ENOSYS ()
113 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115 #define mknod(path,mode,dev) EIO_ENOSYS ()
116 #define sync() EIO_ENOSYS ()
117
118 /* we could even stat and see if it exists */
119 static int
120 symlink (const char *old, const char *neu)
121 {
122 if (CreateSymbolicLink (neu, old, 1))
123 return 0;
124
125 if (CreateSymbolicLink (neu, old, 0))
126 return 0;
127
128 return EIO_ERRNO (ENOENT, -1);
129 }
30 130
31#else 131#else
32 132
33# include "config.h"
34# include <sys/time.h> 133 #include <sys/time.h>
35# include <sys/select.h> 134 #include <sys/select.h>
135 #include <sys/statvfs.h>
36# include <unistd.h> 136 #include <unistd.h>
37# include <utime.h> 137 #include <utime.h>
38# include <signal.h> 138 #include <signal.h>
39# include <dirent.h> 139 #include <dirent.h>
40 140
141 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142 #include <sys/mman.h>
143 #endif
144
145 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
146 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
147 #define _DIRENT_HAVE_D_TYPE /* sigh */
148 #define D_INO(de) (de)->d_fileno
149 #define D_NAMLEN(de) (de)->d_namlen
150 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
151 #define D_INO(de) (de)->d_ino
152 #endif
153
154 #ifdef _D_EXACT_NAMLEN
155 #undef D_NAMLEN
156 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
157 #endif
158
159 #ifdef _DIRENT_HAVE_D_TYPE
160 #define D_TYPE(de) (de)->d_type
161 #endif
162
41# ifndef EIO_STRUCT_DIRENT 163 #ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 164 #define EIO_STRUCT_DIRENT struct dirent
43# endif 165 #endif
44 166
45#endif 167#endif
46 168
47#if HAVE_SENDFILE 169#if HAVE_SENDFILE
48# if __linux 170# if __linux
49# include <sys/sendfile.h> 171# include <sys/sendfile.h>
50# elif __freebsd 172# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 173# include <sys/socket.h>
52# include <sys/uio.h> 174# include <sys/uio.h>
53# elif __hpux 175# elif __hpux
54# include <sys/socket.h> 176# include <sys/socket.h>
55# elif __solaris /* not yet */ 177# elif __solaris
56# include <sys/sendfile.h> 178# include <sys/sendfile.h>
57# else 179# else
58# error sendfile support requested but not available 180# error sendfile support requested but not available
59# endif 181# endif
60#endif 182#endif
61 183
62/* number of seconds after which an idle threads exit */ 184#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 185# define D_TYPE(de) 0
186#endif
187#ifndef D_INO
188# define D_INO(de) 0
189#endif
190#ifndef D_NAMLEN
191# define D_NAMLEN(de) strlen ((de)->d_name)
192#endif
64 193
65/* used for struct dirent, AIX doesn't provide it */ 194/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 195#ifndef NAME_MAX
67# define NAME_MAX 4096 196# define NAME_MAX 4096
68#endif 197#endif
69 198
199/* used for readlink etc. */
200#ifndef PATH_MAX
201# define PATH_MAX 4096
202#endif
203
70/* buffer size for various temporary buffers */ 204/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 205#define EIO_BUFSIZE 65536
72 206
73#define dBUF \ 207#define dBUF \
74 char *eio_buf; \ 208 char *eio_buf; \
75 X_LOCK (wrklock); \ 209 ETP_WORKER_LOCK (self); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 210 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \ 211 ETP_WORKER_UNLOCK (self); \
78 errno = ENOMEM; \ 212 errno = ENOMEM; \
79 if (!eio_buf) \ 213 if (!eio_buf) \
80 return -1; 214 return -1;
81 215
82#define EIO_TICKS ((1000000 + 1023) >> 10) 216#define EIO_TICKS ((1000000 + 1023) >> 10)
83 217
218#define ETP_PRI_MIN EIO_PRI_MIN
219#define ETP_PRI_MAX EIO_PRI_MAX
220
221struct etp_worker;
222
223#define ETP_REQ eio_req
224#define ETP_DESTROY(req) eio_destroy (req)
225static int eio_finish (eio_req *req);
226#define ETP_FINISH(req) eio_finish (req)
227static void eio_execute (struct etp_worker *self, eio_req *req);
228#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
229
230#define ETP_WORKER_CLEAR(req) \
231 if (wrk->dbuf) \
232 { \
233 free (wrk->dbuf); \
234 wrk->dbuf = 0; \
235 } \
236 \
237 if (wrk->dirp) \
238 { \
239 closedir (wrk->dirp); \
240 wrk->dirp = 0; \
241 }
242
243#define ETP_WORKER_COMMON \
244 void *dbuf; \
245 DIR *dirp;
246
247/*****************************************************************************/
248
249#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
250
251/* calculate time difference in ~1/EIO_TICKS of a second */
252ecb_inline int
253tvdiff (struct timeval *tv1, struct timeval *tv2)
254{
255 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
256 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
257}
258
259static unsigned int started, idle, wanted = 4;
260
84static void (*want_poll_cb) (void); 261static void (*want_poll_cb) (void);
85static void (*done_poll_cb) (void); 262static void (*done_poll_cb) (void);
86 263
87static unsigned int max_poll_time = 0; 264static unsigned int max_poll_time; /* reslock */
88static unsigned int max_poll_reqs = 0; 265static unsigned int max_poll_reqs; /* reslock */
89 266
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 267static volatile unsigned int nreqs; /* reqlock */
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 268static volatile unsigned int nready; /* reqlock */
92{ 269static volatile unsigned int npending; /* reqlock */
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95}
96 272
97static unsigned int started, idle, wanted = 4; 273static xmutex_t wrklock;
274static xmutex_t reslock;
275static xmutex_t reqlock;
276static xcond_t reqwait;
277
278#if !HAVE_PREADWRITE
279/*
280 * make our pread/pwrite emulation safe against themselves, but not against
281 * normal read/write by using a mutex. slows down execution a lot,
282 * but that's your problem, not mine.
283 */
284static xmutex_t preadwritelock = X_MUTEX_INIT;
285#endif
286
287typedef struct etp_worker
288{
289 /* locked by wrklock */
290 struct etp_worker *prev, *next;
291
292 xthread_t tid;
293
294 /* locked by reslock, reqlock or wrklock */
295 ETP_REQ *req; /* currently processed request */
296
297 ETP_WORKER_COMMON
298} etp_worker;
299
300static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
301
302#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
303#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98 304
99/* worker threads management */ 305/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101 306
102typedef struct worker 307static void ecb_cold
103{
104 /* locked by wrklock */
105 struct worker *prev, *next;
106
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk) 308etp_worker_clear (etp_worker *wrk)
118{ 309{
119 if (wrk->dirp) 310 ETP_WORKER_CLEAR (wrk);
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124
125 if (wrk->dbuf)
126 {
127 free (wrk->dbuf);
128 wrk->dbuf = 0;
129 }
130} 311}
131 312
132static void worker_free (worker *wrk) 313static void ecb_cold
314etp_worker_free (etp_worker *wrk)
133{ 315{
134 wrk->next->prev = wrk->prev; 316 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next; 317 wrk->prev->next = wrk->next;
136 318
137 free (wrk); 319 free (wrk);
138} 320}
139 321
140static volatile unsigned int nreqs, nready, npending; 322static unsigned int
141static volatile unsigned int max_idle = 4; 323etp_nreqs (void)
142
143static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT;
146
147unsigned int eio_nreqs (void)
148{ 324{
325 int retval;
326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
327 retval = nreqs;
328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149 return nreqs; 329 return retval;
150} 330}
151 331
152unsigned int eio_nready (void) 332static unsigned int
333etp_nready (void)
153{ 334{
154 unsigned int retval; 335 unsigned int retval;
155 336
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready; 338 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159 340
160 return retval; 341 return retval;
161} 342}
162 343
163unsigned int eio_npending (void) 344static unsigned int
345etp_npending (void)
164{ 346{
165 unsigned int retval; 347 unsigned int retval;
166 348
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending; 350 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170 352
171 return retval; 353 return retval;
172} 354}
173 355
174unsigned int eio_nthreads (void) 356static unsigned int
357etp_nthreads (void)
175{ 358{
176 unsigned int retval; 359 unsigned int retval;
177 360
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started; 362 retval = started;
… …
186 * a somewhat faster data structure might be nice, but 369 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns 370 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation. 371 * per shift, the most expensive operation.
189 */ 372 */
190typedef struct { 373typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 374 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192 int size; 375 int size;
193} reqq; 376} etp_reqq;
194 377
195static reqq req_queue; 378static etp_reqq req_queue;
196static reqq res_queue; 379static etp_reqq res_queue;
197 380
198static int reqq_push (reqq *q, eio_req *req) 381static int ecb_noinline
382reqq_push (etp_reqq *q, ETP_REQ *req)
199{ 383{
200 int pri = req->pri; 384 int pri = req->pri;
201 req->next = 0; 385 req->next = 0;
202 386
203 if (q->qe[pri]) 387 if (q->qe[pri])
… …
209 q->qe[pri] = q->qs[pri] = req; 393 q->qe[pri] = q->qs[pri] = req;
210 394
211 return q->size++; 395 return q->size++;
212} 396}
213 397
214static eio_req *reqq_shift (reqq *q) 398static ETP_REQ * ecb_noinline
399reqq_shift (etp_reqq *q)
215{ 400{
216 int pri; 401 int pri;
217 402
218 if (!q->size) 403 if (!q->size)
219 return 0; 404 return 0;
220 405
221 --q->size; 406 --q->size;
222 407
223 for (pri = EIO_NUM_PRI; pri--; ) 408 for (pri = ETP_NUM_PRI; pri--; )
224 { 409 {
225 eio_req *req = q->qs[pri]; 410 eio_req *req = q->qs[pri];
226 411
227 if (req) 412 if (req)
228 { 413 {
… …
234 } 419 }
235 420
236 abort (); 421 abort ();
237} 422}
238 423
239static void grp_try_feed (eio_req *grp) 424static void ecb_cold
425etp_thread_init (void)
240{ 426{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 427#if !HAVE_PREADWRITE
428 X_MUTEX_CREATE (preadwritelock);
429#endif
430 X_MUTEX_CREATE (wrklock);
431 X_MUTEX_CREATE (reslock);
432 X_MUTEX_CREATE (reqlock);
433 X_COND_CREATE (reqwait);
434}
435
436static void ecb_cold
437etp_atfork_prepare (void)
438{
439}
440
441static void ecb_cold
442etp_atfork_parent (void)
443{
444}
445
446static void ecb_cold
447etp_atfork_child (void)
448{
449 ETP_REQ *prv;
450
451 while ((prv = reqq_shift (&req_queue)))
452 ETP_DESTROY (prv);
453
454 while ((prv = reqq_shift (&res_queue)))
455 ETP_DESTROY (prv);
456
457 while (wrk_first.next != &wrk_first)
242 { 458 {
243 int old_len = grp->size; 459 etp_worker *wrk = wrk_first.next;
244 460
245 EIO_FEED (grp); 461 if (wrk->req)
462 ETP_DESTROY (wrk->req);
246 463
247 /* stop if no progress has been made */ 464 etp_worker_clear (wrk);
248 if (old_len == grp->size) 465 etp_worker_free (wrk);
249 {
250 grp->feed = 0;
251 break;
252 }
253 } 466 }
254}
255 467
256static int eio_finish (eio_req *req); 468 started = 0;
469 idle = 0;
470 nreqs = 0;
471 nready = 0;
472 npending = 0;
257 473
258static int grp_dec (eio_req *grp) 474 etp_thread_init ();
259{ 475}
260 --grp->size;
261 476
262 /* call feeder, if applicable */ 477static void ecb_cold
263 grp_try_feed (grp); 478etp_once_init (void)
479{
480 etp_thread_init ();
481 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
482}
264 483
265 /* finish, if done */ 484static int ecb_cold
266 if (!grp->size && grp->int1) 485etp_init (void (*want_poll)(void), void (*done_poll)(void))
267 return eio_finish (grp); 486{
268 else 487 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
488
489 pthread_once (&doinit, etp_once_init);
490
491 want_poll_cb = want_poll;
492 done_poll_cb = done_poll;
493
269 return 0; 494 return 0;
270} 495}
271 496
272void eio_destroy (eio_req *req)
273{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277 EIO_DESTROY (req);
278}
279
280static int eio_finish (eio_req *req)
281{
282 int res = EIO_FINISH (req);
283
284 if (req->grp)
285 {
286 int res2;
287 eio_req *grp = req->grp;
288
289 /* unlink request */
290 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next;
295
296 res2 = grp_dec (grp);
297
298 if (!res && res2)
299 res = res2;
300 }
301
302 eio_destroy (req);
303
304 return res;
305}
306
307void eio_grp_cancel (eio_req *grp)
308{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next)
310 eio_cancel (grp);
311}
312
313void eio_cancel (eio_req *req)
314{
315 X_LOCK (wrklock);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 497X_THREAD_PROC (etp_proc);
323 498
324static void start_thread (void) 499static void ecb_cold
500etp_start_thread (void)
325{ 501{
326 worker *wrk = calloc (1, sizeof (worker)); 502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 503
328 /*TODO*/ 504 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk)); 505 assert (("unable to allocate worker thread data", wrk));
330 506
331 X_LOCK (wrklock); 507 X_LOCK (wrklock);
332 508
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 509 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 510 {
335 wrk->prev = &wrk_first; 511 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 512 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 513 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 514 wrk_first.next = wrk;
… …
342 free (wrk); 518 free (wrk);
343 519
344 X_UNLOCK (wrklock); 520 X_UNLOCK (wrklock);
345} 521}
346 522
523static void
347static void maybe_start_thread (void) 524etp_maybe_start_thread (void)
348{ 525{
349 if (eio_nthreads () >= wanted) 526 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 527 return;
351 528
352 /* todo: maybe use idle here, but might be less exact */ 529 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 530 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 531 return;
355 532
356 start_thread (); 533 etp_start_thread ();
357} 534}
358 535
359void eio_submit (eio_req *req) 536static void ecb_cold
360{ 537etp_end_thread (void)
361 req->pri += EIO_PRI_BIAS;
362
363 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366 ++nreqs;
367
368 X_LOCK (reqlock);
369 ++nready;
370 reqq_push (&req_queue, req);
371 X_COND_SIGNAL (reqwait);
372 X_UNLOCK (reqlock);
373
374 maybe_start_thread ();
375}
376
377static void end_thread (void)
378{ 538{
379 eio_req *req = calloc (1, sizeof (eio_req)); 539 eio_req *req = calloc (1, sizeof (eio_req));
380 540
381 req->type = EIO_QUIT; 541 req->type = -1;
382 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383 543
384 X_LOCK (reqlock); 544 X_LOCK (reqlock);
385 reqq_push (&req_queue, req); 545 reqq_push (&req_queue, req);
386 X_COND_SIGNAL (reqwait); 546 X_COND_SIGNAL (reqwait);
387 X_UNLOCK (reqlock); 547 X_UNLOCK (reqlock);
… …
389 X_LOCK (wrklock); 549 X_LOCK (wrklock);
390 --started; 550 --started;
391 X_UNLOCK (wrklock); 551 X_UNLOCK (wrklock);
392} 552}
393 553
394void eio_set_max_poll_time (double nseconds) 554static int
395{
396 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397 max_poll_time = nseconds;
398 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399}
400
401void eio_set_max_poll_reqs (unsigned int maxreqs)
402{
403 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404 max_poll_reqs = maxreqs;
405 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406}
407
408void eio_set_max_idle (unsigned int nthreads)
409{
410 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411 max_idle = nthreads <= 0 ? 1 : nthreads;
412 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413}
414
415void eio_set_min_parallel (unsigned int nthreads)
416{
417 if (wanted < nthreads)
418 wanted = nthreads;
419}
420
421void eio_set_max_parallel (unsigned int nthreads)
422{
423 if (wanted > nthreads)
424 wanted = nthreads;
425
426 while (started > wanted)
427 end_thread ();
428}
429
430int eio_poll (void) 555etp_poll (void)
431{ 556{
432 int maxreqs = max_poll_reqs; 557 unsigned int maxreqs;
558 unsigned int maxtime;
433 struct timeval tv_start, tv_now; 559 struct timeval tv_start, tv_now;
434 eio_req *req;
435 560
561 X_LOCK (reslock);
562 maxreqs = max_poll_reqs;
563 maxtime = max_poll_time;
564 X_UNLOCK (reslock);
565
436 if (max_poll_time) 566 if (maxtime)
437 gettimeofday (&tv_start, 0); 567 gettimeofday (&tv_start, 0);
438 568
439 for (;;) 569 for (;;)
440 { 570 {
571 ETP_REQ *req;
572
441 maybe_start_thread (); 573 etp_maybe_start_thread ();
442 574
443 X_LOCK (reslock); 575 X_LOCK (reslock);
444 req = reqq_shift (&res_queue); 576 req = reqq_shift (&res_queue);
445 577
446 if (req) 578 if (req)
… …
454 X_UNLOCK (reslock); 586 X_UNLOCK (reslock);
455 587
456 if (!req) 588 if (!req)
457 return 0; 589 return 0;
458 590
591 X_LOCK (reqlock);
459 --nreqs; 592 --nreqs;
593 X_UNLOCK (reqlock);
460 594
461 if (req->type == EIO_GROUP && req->size) 595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
462 { 596 {
463 req->int1 = 1; /* mark request as delayed */ 597 req->int1 = 1; /* mark request as delayed */
464 continue; 598 continue;
465 } 599 }
466 else 600 else
467 { 601 {
468 int res = eio_finish (req); 602 int res = ETP_FINISH (req);
469 if (res) 603 if (ecb_expect_false (res))
470 return res; 604 return res;
471 } 605 }
472 606
473 if (maxreqs && !--maxreqs) 607 if (ecb_expect_false (maxreqs && !--maxreqs))
474 break; 608 break;
475 609
476 if (max_poll_time) 610 if (maxtime)
477 { 611 {
478 gettimeofday (&tv_now, 0); 612 gettimeofday (&tv_now, 0);
479 613
480 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 614 if (tvdiff (&tv_start, &tv_now) >= maxtime)
481 break; 615 break;
482 } 616 }
483 } 617 }
484 618
485 errno = EAGAIN; 619 errno = EAGAIN;
486 return -1; 620 return -1;
487} 621}
488 622
623static void
624etp_cancel (ETP_REQ *req)
625{
626 req->cancelled = 1;
627
628 eio_grp_cancel (req);
629}
630
631static void
632etp_submit (ETP_REQ *req)
633{
634 req->pri -= ETP_PRI_MIN;
635
636 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
637 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
638
639 if (ecb_expect_false (req->type == EIO_GROUP))
640 {
641 /* I hope this is worth it :/ */
642 X_LOCK (reqlock);
643 ++nreqs;
644 X_UNLOCK (reqlock);
645
646 X_LOCK (reslock);
647
648 ++npending;
649
650 if (!reqq_push (&res_queue, req) && want_poll_cb)
651 want_poll_cb ();
652
653 X_UNLOCK (reslock);
654 }
655 else
656 {
657 X_LOCK (reqlock);
658 ++nreqs;
659 ++nready;
660 reqq_push (&req_queue, req);
661 X_COND_SIGNAL (reqwait);
662 X_UNLOCK (reqlock);
663
664 etp_maybe_start_thread ();
665 }
666}
667
668static void ecb_cold
669etp_set_max_poll_time (double nseconds)
670{
671 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
672 max_poll_time = nseconds * EIO_TICKS;
673 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
674}
675
676static void ecb_cold
677etp_set_max_poll_reqs (unsigned int maxreqs)
678{
679 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
680 max_poll_reqs = maxreqs;
681 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
682}
683
684static void ecb_cold
685etp_set_max_idle (unsigned int nthreads)
686{
687 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
688 max_idle = nthreads;
689 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
690}
691
692static void ecb_cold
693etp_set_idle_timeout (unsigned int seconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
696 idle_timeout = seconds;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
698}
699
700static void ecb_cold
701etp_set_min_parallel (unsigned int nthreads)
702{
703 if (wanted < nthreads)
704 wanted = nthreads;
705}
706
707static void ecb_cold
708etp_set_max_parallel (unsigned int nthreads)
709{
710 if (wanted > nthreads)
711 wanted = nthreads;
712
713 while (started > wanted)
714 etp_end_thread ();
715}
716
717/*****************************************************************************/
718
719static void
720grp_try_feed (eio_req *grp)
721{
722 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
723 {
724 grp->flags &= ~EIO_FLAG_GROUPADD;
725
726 EIO_FEED (grp);
727
728 /* stop if no progress has been made */
729 if (!(grp->flags & EIO_FLAG_GROUPADD))
730 {
731 grp->feed = 0;
732 break;
733 }
734 }
735}
736
737static int
738grp_dec (eio_req *grp)
739{
740 --grp->size;
741
742 /* call feeder, if applicable */
743 grp_try_feed (grp);
744
745 /* finish, if done */
746 if (!grp->size && grp->int1)
747 return eio_finish (grp);
748 else
749 return 0;
750}
751
752static void
753eio_destroy (eio_req *req)
754{
755 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
756 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
757
758 EIO_DESTROY (req);
759}
760
761static int
762eio_finish (eio_req *req)
763{
764 int res = EIO_FINISH (req);
765
766 if (req->grp)
767 {
768 int res2;
769 eio_req *grp = req->grp;
770
771 /* unlink request */
772 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
773 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
774
775 if (grp->grp_first == req)
776 grp->grp_first = req->grp_next;
777
778 res2 = grp_dec (grp);
779
780 if (!res)
781 res = res2;
782 }
783
784 eio_destroy (req);
785
786 return res;
787}
788
789void
790eio_grp_cancel (eio_req *grp)
791{
792 for (grp = grp->grp_first; grp; grp = grp->grp_next)
793 eio_cancel (grp);
794}
795
796void
797eio_cancel (eio_req *req)
798{
799 etp_cancel (req);
800}
801
802void
803eio_submit (eio_req *req)
804{
805 etp_submit (req);
806}
807
808unsigned int
809eio_nreqs (void)
810{
811 return etp_nreqs ();
812}
813
814unsigned int
815eio_nready (void)
816{
817 return etp_nready ();
818}
819
820unsigned int
821eio_npending (void)
822{
823 return etp_npending ();
824}
825
826unsigned int ecb_cold
827eio_nthreads (void)
828{
829 return etp_nthreads ();
830}
831
832void ecb_cold
833eio_set_max_poll_time (double nseconds)
834{
835 etp_set_max_poll_time (nseconds);
836}
837
838void ecb_cold
839eio_set_max_poll_reqs (unsigned int maxreqs)
840{
841 etp_set_max_poll_reqs (maxreqs);
842}
843
844void ecb_cold
845eio_set_max_idle (unsigned int nthreads)
846{
847 etp_set_max_idle (nthreads);
848}
849
850void ecb_cold
851eio_set_idle_timeout (unsigned int seconds)
852{
853 etp_set_idle_timeout (seconds);
854}
855
856void ecb_cold
857eio_set_min_parallel (unsigned int nthreads)
858{
859 etp_set_min_parallel (nthreads);
860}
861
862void ecb_cold
863eio_set_max_parallel (unsigned int nthreads)
864{
865 etp_set_max_parallel (nthreads);
866}
867
868int eio_poll (void)
869{
870 return etp_poll ();
871}
872
489/*****************************************************************************/ 873/*****************************************************************************/
490/* work around various missing functions */ 874/* work around various missing functions */
491 875
492#if !HAVE_PREADWRITE 876#if !HAVE_PREADWRITE
877# undef pread
878# undef pwrite
493# define pread eio__pread 879# define pread eio__pread
494# define pwrite eio__pwrite 880# define pwrite eio__pwrite
495 881
496/*
497 * make our pread/pwrite safe against themselves, but not against
498 * normal read/write by using a mutex. slows down execution a lot,
499 * but that's your problem, not mine.
500 */
501static mutex_t preadwritelock = X_MUTEX_INIT;
502
503static ssize_t 882static eio_ssize_t
504eio__pread (int fd, void *buf, size_t count, off_t offset) 883eio__pread (int fd, void *buf, size_t count, off_t offset)
505{ 884{
506 ssize_t res; 885 eio_ssize_t res;
507 off_t ooffset; 886 off_t ooffset;
508 887
509 X_LOCK (preadwritelock); 888 X_LOCK (preadwritelock);
510 ooffset = lseek (fd, 0, SEEK_CUR); 889 ooffset = lseek (fd, 0, SEEK_CUR);
511 lseek (fd, offset, SEEK_SET); 890 lseek (fd, offset, SEEK_SET);
… …
514 X_UNLOCK (preadwritelock); 893 X_UNLOCK (preadwritelock);
515 894
516 return res; 895 return res;
517} 896}
518 897
519static ssize_t 898static eio_ssize_t
520eio__pwrite (int fd, void *buf, size_t count, off_t offset) 899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521{ 900{
522 ssize_t res; 901 eio_ssize_t res;
523 off_t ooffset; 902 off_t ooffset;
524 903
525 X_LOCK (preadwritelock); 904 X_LOCK (preadwritelock);
526 ooffset = lseek (fd, 0, SEEK_CUR); 905 ooffset = lseek (fd, 0, SEEK_CUR);
527 lseek (fd, offset, SEEK_SET); 906 lseek (fd, offset, SEEK_SET);
528 res = write (fd, buf, count); 907 res = write (fd, buf, count);
529 lseek (fd, offset, SEEK_SET); 908 lseek (fd, ooffset, SEEK_SET);
530 X_UNLOCK (preadwritelock); 909 X_UNLOCK (preadwritelock);
531 910
532 return res; 911 return res;
533} 912}
534#endif 913#endif
535 914
536#ifndef HAVE_FUTIMES 915#ifndef HAVE_UTIMES
537 916
917# undef utimes
538# define utimes(path,times) eio__utimes (path, times) 918# define utimes(path,times) eio__utimes (path, times)
539# define futimes(fd,times) eio__futimes (fd, times)
540 919
541static int 920static int
542eio__utimes (const char *filename, const struct timeval times[2]) 921eio__utimes (const char *filename, const struct timeval times[2])
543{ 922{
544 if (times) 923 if (times)
… …
552 } 931 }
553 else 932 else
554 return utime (filename, 0); 933 return utime (filename, 0);
555} 934}
556 935
936#endif
937
938#ifndef HAVE_FUTIMES
939
940# undef futimes
941# define futimes(fd,times) eio__futimes (fd, times)
942
943static int
557static int eio__futimes (int fd, const struct timeval tv[2]) 944eio__futimes (int fd, const struct timeval tv[2])
558{ 945{
559 errno = ENOSYS; 946 errno = ENOSYS;
560 return -1; 947 return -1;
561} 948}
562 949
563#endif 950#endif
564 951
565#if !HAVE_FDATASYNC 952#if !HAVE_FDATASYNC
953# undef fdatasync
566# define fdatasync fsync 954# define fdatasync(fd) fsync (fd)
567#endif 955#endif
956
957/* sync_file_range always needs emulation */
958static int
959eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
960{
961#if HAVE_SYNC_FILE_RANGE
962 int res;
963
964 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
965 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
966 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
967 {
968 flags = 0
969 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
970 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
971 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
972 }
973
974 res = sync_file_range (fd, offset, nbytes, flags);
975
976 if (!res || errno != ENOSYS)
977 return res;
978#endif
979
980 /* even though we could play tricks with the flags, it's better to always
981 * call fdatasync, as that matches the expectation of its users best */
982 return fdatasync (fd);
983}
984
985static int
986eio__fallocate (int fd, int mode, off_t offset, size_t len)
987{
988#if HAVE_FALLOCATE
989 return fallocate (fd, mode, offset, len);
990#else
991 errno = ENOSYS;
992 return -1;
993#endif
994}
568 995
569#if !HAVE_READAHEAD 996#if !HAVE_READAHEAD
997# undef readahead
570# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 998# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571 999
572static ssize_t 1000static eio_ssize_t
573eio__readahead (int fd, off_t offset, size_t count, worker *self) 1001eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574{ 1002{
575 size_t todo = count; 1003 size_t todo = count;
576 dBUF; 1004 dBUF;
577 1005
578 while (todo > 0) 1006 while (todo > 0)
… …
588 return count; 1016 return count;
589} 1017}
590 1018
591#endif 1019#endif
592 1020
593#if !HAVE_READDIR_R
594# define readdir_r eio__readdir_r
595
596static mutex_t readdirlock = X_MUTEX_INIT;
597
598static int
599eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600{
601 EIO_STRUCT_DIRENT *e;
602 int errorno;
603
604 X_LOCK (readdirlock);
605
606 e = readdir (dirp);
607 errorno = errno;
608
609 if (e)
610 {
611 *res = ent;
612 strcpy (ent->d_name, e->d_name);
613 }
614 else
615 *res = 0;
616
617 X_UNLOCK (readdirlock);
618
619 errno = errorno;
620 return e ? 0 : -1;
621}
622#endif
623
624/* sendfile always needs emulation */ 1021/* sendfile always needs emulation */
625static ssize_t 1022static eio_ssize_t
626eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 1023eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
627{ 1024{
1025 eio_ssize_t written = 0;
628 ssize_t res; 1026 eio_ssize_t res;
629 1027
630 if (!count) 1028 if (!count)
631 return 0; 1029 return 0;
632 1030
1031 for (;;)
1032 {
1033#ifdef __APPLE__
1034# undef HAVE_SENDFILE /* broken, as everything on os x */
1035#endif
633#if HAVE_SENDFILE 1036#if HAVE_SENDFILE
634# if __linux 1037# if __linux
1038 off_t soffset = offset;
635 res = sendfile (ofd, ifd, &offset, count); 1039 res = sendfile (ofd, ifd, &soffset, count);
636 1040
637# elif __freebsd 1041# elif __FreeBSD__
638 /* 1042 /*
639 * Of course, the freebsd sendfile is a dire hack with no thoughts 1043 * Of course, the freebsd sendfile is a dire hack with no thoughts
640 * wasted on making it similar to other I/O functions. 1044 * wasted on making it similar to other I/O functions.
641 */ 1045 */
642 {
643 off_t sbytes; 1046 off_t sbytes;
644 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1047 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645 1048
646 if (res < 0 && sbytes) 1049 #if 0 /* according to the manpage, this is correct, but broken behaviour */
647 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1050 /* freebsd' sendfile will return 0 on success */
1051 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1052 /* not on e.g. EIO or EPIPE - sounds broken */
1053 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648 res = sbytes; 1054 res = sbytes;
649 } 1055 #endif
1056
1057 /* according to source inspection, this is correct, and useful behaviour */
1058 if (sbytes)
1059 res = sbytes;
1060
1061# elif defined (__APPLE__)
1062 off_t sbytes = count;
1063 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064
1065 /* according to the manpage, sbytes is always valid */
1066 if (sbytes)
1067 res = sbytes;
650 1068
651# elif __hpux 1069# elif __hpux
652 res = sendfile (ofd, ifd, offset, count, 0, 0); 1070 res = sendfile (ofd, ifd, offset, count, 0, 0);
653 1071
654# elif __solaris 1072# elif __solaris
655 {
656 struct sendfilevec vec; 1073 struct sendfilevec vec;
657 size_t sbytes; 1074 size_t sbytes;
658 1075
659 vec.sfv_fd = ifd; 1076 vec.sfv_fd = ifd;
660 vec.sfv_flag = 0; 1077 vec.sfv_flag = 0;
661 vec.sfv_off = offset; 1078 vec.sfv_off = offset;
662 vec.sfv_len = count; 1079 vec.sfv_len = count;
663 1080
664 res = sendfilev (ofd, &vec, 1, &sbytes); 1081 res = sendfilev (ofd, &vec, 1, &sbytes);
665 1082
666 if (res < 0 && sbytes) 1083 if (res < 0 && sbytes)
667 res = sbytes; 1084 res = sbytes;
668 }
669 1085
670# endif 1086# endif
1087
1088#elif defined (_WIN32) && 0
1089 /* does not work, just for documentation of what would need to be done */
1090 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1091 /* libeio guarantees that the file offset does not change, and windows */
1092 /* has no way to get an independent handle to the same file description */
1093 HANDLE h = TO_SOCKET (ifd);
1094 SetFilePointer (h, offset, 0, FILE_BEGIN);
1095 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096
671#else 1097#else
672 res = -1; 1098 res = -1;
673 errno = ENOSYS; 1099 errno = ENOSYS;
674#endif 1100#endif
675 1101
1102 /* we assume sendfile can copy at least 128mb in one go */
1103 if (res <= 128 * 1024 * 1024)
1104 {
1105 if (res > 0)
1106 written += res;
1107
1108 if (written)
1109 return written;
1110
1111 break;
1112 }
1113 else
1114 {
1115 /* if we requested more, then probably the kernel was lazy */
1116 written += res;
1117 offset += res;
1118 count -= res;
1119
1120 if (!count)
1121 return written;
1122 }
1123 }
1124
676 if (res < 0 1125 if (res < 0
677 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1126 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127 /* BSDs */
1128#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1129 || errno == ENOTSUP
1130#endif
1131 || errno == EOPNOTSUPP /* BSDs */
678#if __solaris 1132#if __solaris
679 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
680#endif 1134#endif
681 ) 1135 )
682 ) 1136 )
… …
686 1140
687 res = 0; 1141 res = 0;
688 1142
689 while (count) 1143 while (count)
690 { 1144 {
691 ssize_t cnt; 1145 eio_ssize_t cnt;
692 1146
693 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1147 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694 1148
695 if (cnt <= 0) 1149 if (cnt <= 0)
696 { 1150 {
… …
713 } 1167 }
714 1168
715 return res; 1169 return res;
716} 1170}
717 1171
1172#ifdef PAGESIZE
1173# define eio_pagesize() PAGESIZE
1174#else
1175static intptr_t
1176eio_pagesize (void)
1177{
1178 static intptr_t page;
1179
1180 if (!page)
1181 page = sysconf (_SC_PAGESIZE);
1182
1183 return page;
1184}
1185#endif
1186
1187static void
1188eio_page_align (void **addr, size_t *length)
1189{
1190 intptr_t mask = eio_pagesize () - 1;
1191
1192 /* round down addr */
1193 intptr_t adj = mask & (intptr_t)*addr;
1194
1195 *addr = (void *)((intptr_t)*addr - adj);
1196 *length += adj;
1197
1198 /* round up length */
1199 *length = (*length + mask) & ~mask;
1200}
1201
1202#if !_POSIX_MEMLOCK
1203# define eio__mlockall(a) EIO_ENOSYS ()
1204#else
1205
1206static int
1207eio__mlockall (int flags)
1208{
1209 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1210 extern int mallopt (int, int);
1211 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1212 #endif
1213
1214 if (EIO_MCL_CURRENT != MCL_CURRENT
1215 || EIO_MCL_FUTURE != MCL_FUTURE)
1216 {
1217 flags = 0
1218 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1219 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1220 }
1221
1222 return mlockall (flags);
1223}
1224#endif
1225
1226#if !_POSIX_MEMLOCK_RANGE
1227# define eio__mlock(a,b) EIO_ENOSYS ()
1228#else
1229
1230static int
1231eio__mlock (void *addr, size_t length)
1232{
1233 eio_page_align (&addr, &length);
1234
1235 return mlock (addr, length);
1236}
1237
1238#endif
1239
1240#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1241# define eio__msync(a,b,c) EIO_ENOSYS ()
1242#else
1243
1244static int
1245eio__msync (void *mem, size_t len, int flags)
1246{
1247 eio_page_align (&mem, &len);
1248
1249 if (EIO_MS_ASYNC != MS_SYNC
1250 || EIO_MS_INVALIDATE != MS_INVALIDATE
1251 || EIO_MS_SYNC != MS_SYNC)
1252 {
1253 flags = 0
1254 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1255 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1256 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1257 }
1258
1259 return msync (mem, len, flags);
1260}
1261
1262#endif
1263
1264static int
1265eio__mtouch (eio_req *req)
1266{
1267 void *mem = req->ptr2;
1268 size_t len = req->size;
1269 int flags = req->int1;
1270
1271 eio_page_align (&mem, &len);
1272
1273 {
1274 intptr_t addr = (intptr_t)mem;
1275 intptr_t end = addr + len;
1276 intptr_t page = eio_pagesize ();
1277
1278 if (addr < end)
1279 if (flags & EIO_MT_MODIFY) /* modify */
1280 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1281 else
1282 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1283 }
1284
1285 return 0;
1286}
1287
1288/*****************************************************************************/
1289/* requests implemented outside eio_execute, because they are so large */
1290
1291static void
1292eio__realpath (eio_req *req, etp_worker *self)
1293{
1294 char *rel = req->ptr1;
1295 char *res;
1296 char *tmp1, *tmp2;
1297#if SYMLOOP_MAX > 32
1298 int symlinks = SYMLOOP_MAX;
1299#else
1300 int symlinks = 32;
1301#endif
1302
1303 req->result = -1;
1304
1305 errno = EINVAL;
1306 if (!rel)
1307 return;
1308
1309 errno = ENOENT;
1310 if (!*rel)
1311 return;
1312
1313 if (!req->ptr2)
1314 {
1315 X_LOCK (wrklock);
1316 req->flags |= EIO_FLAG_PTR2_FREE;
1317 X_UNLOCK (wrklock);
1318 req->ptr2 = malloc (PATH_MAX * 3);
1319
1320 errno = ENOMEM;
1321 if (!req->ptr2)
1322 return;
1323 }
1324
1325 res = req->ptr2;
1326 tmp1 = res + PATH_MAX;
1327 tmp2 = tmp1 + PATH_MAX;
1328
1329#if 0 /* disabled, the musl way to do things is just too racy */
1330#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1331 /* on linux we may be able to ask the kernel */
1332 {
1333 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1334
1335 if (fd >= 0)
1336 {
1337 sprintf (tmp1, "/proc/self/fd/%d", fd);
1338 req->result = readlink (tmp1, res, PATH_MAX);
1339 close (fd);
1340
1341 /* here we should probably stat the open file and the disk file, to make sure they still match */
1342
1343 if (req->result > 0)
1344 goto done;
1345 }
1346 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1347 return;
1348 }
1349#endif
1350#endif
1351
1352 if (*rel != '/')
1353 {
1354 if (!getcwd (res, PATH_MAX))
1355 return;
1356
1357 if (res [1]) /* only use if not / */
1358 res += strlen (res);
1359 }
1360
1361 while (*rel)
1362 {
1363 eio_ssize_t len, linklen;
1364 char *beg = rel;
1365
1366 while (*rel && *rel != '/')
1367 ++rel;
1368
1369 len = rel - beg;
1370
1371 if (!len) /* skip slashes */
1372 {
1373 ++rel;
1374 continue;
1375 }
1376
1377 if (beg [0] == '.')
1378 {
1379 if (len == 1)
1380 continue; /* . - nop */
1381
1382 if (beg [1] == '.' && len == 2)
1383 {
1384 /* .. - back up one component, if possible */
1385
1386 while (res != req->ptr2)
1387 if (*--res == '/')
1388 break;
1389
1390 continue;
1391 }
1392 }
1393
1394 errno = ENAMETOOLONG;
1395 if (res + 1 + len + 1 >= tmp1)
1396 return;
1397
1398 /* copy one component */
1399 *res = '/';
1400 memcpy (res + 1, beg, len);
1401
1402 /* zero-terminate, for readlink */
1403 res [len + 1] = 0;
1404
1405 /* now check if it's a symlink */
1406 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1407
1408 if (linklen < 0)
1409 {
1410 if (errno != EINVAL)
1411 return;
1412
1413 /* it's a normal directory. hopefully */
1414 res += len + 1;
1415 }
1416 else
1417 {
1418 /* yay, it was a symlink - build new path in tmp2 */
1419 int rellen = strlen (rel);
1420
1421 errno = ENAMETOOLONG;
1422 if (linklen + 1 + rellen >= PATH_MAX)
1423 return;
1424
1425 errno = ELOOP;
1426 if (!--symlinks)
1427 return;
1428
1429 if (*tmp1 == '/')
1430 res = req->ptr2; /* symlink resolves to an absolute path */
1431
1432 /* we need to be careful, as rel might point into tmp2 already */
1433 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434 tmp2 [linklen] = '/';
1435 memcpy (tmp2, tmp1, linklen);
1436
1437 rel = tmp2;
1438 }
1439 }
1440
1441 /* special case for the lone root path */
1442 if (res == req->ptr2)
1443 *res++ = '/';
1444
1445 req->result = res - (char *)req->ptr2;
1446
1447done:
1448 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449}
1450
1451static signed char
1452eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1453{
1454 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1455 : a->inode < b->inode ? -1
1456 : a->inode > b->inode ? 1
1457 : 0;
1458}
1459
1460#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1461
1462#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1463#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464
1465static void
1466eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1467{
1468 unsigned char bits [9 + sizeof (ino_t) * 8];
1469 unsigned char *bit = bits;
1470
1471 assert (CHAR_BIT == 8);
1472 assert (sizeof (eio_dirent) * 8 < 256);
1473 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1474 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1475
1476 if (size <= EIO_SORT_FAST)
1477 return;
1478
1479 /* first prepare an array of bits to test in our radix sort */
1480 /* try to take endianness into account, as well as differences in ino_t sizes */
1481 /* inode_bits must contain all inodes ORed together */
1482 /* which is used to skip bits that are 0 everywhere, which is very common */
1483 {
1484 ino_t endianness;
1485 int i, j;
1486
1487 /* we store the byte offset of byte n into byte n of "endianness" */
1488 for (i = 0; i < sizeof (ino_t); ++i)
1489 ((unsigned char *)&endianness)[i] = i;
1490
1491 *bit++ = 0;
1492
1493 for (i = 0; i < sizeof (ino_t); ++i)
1494 {
1495 /* shifting off the byte offsets out of "endianness" */
1496 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497 endianness >>= 8;
1498
1499 for (j = 0; j < 8; ++j)
1500 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1501 *bit++ = offs + j;
1502 }
1503
1504 for (j = 0; j < 8; ++j)
1505 if (score_bits & (1 << j))
1506 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1507 }
1508
1509 /* now actually do the sorting (a variant of MSD radix sort) */
1510 {
1511 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1512 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1513 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1514 int stk_idx = 0;
1515
1516 base_stk [stk_idx] = dents;
1517 end_stk [stk_idx] = dents + size;
1518 bit_stk [stk_idx] = bit - 1;
1519
1520 do
1521 {
1522 base = base_stk [stk_idx];
1523 end = end_stk [stk_idx];
1524 bit = bit_stk [stk_idx];
1525
1526 for (;;)
1527 {
1528 unsigned char O = *bit >> 3;
1529 unsigned char M = 1 << (*bit & 7);
1530
1531 eio_dirent *a = base;
1532 eio_dirent *b = end;
1533
1534 if (b - a < EIO_SORT_CUTOFF)
1535 break;
1536
1537 /* now bit-partition the array on the bit */
1538 /* this ugly asymmetric loop seems to perform much better than typical */
1539 /* partition algos found in the literature */
1540 do
1541 if (!(((unsigned char *)a)[O] & M))
1542 ++a;
1543 else if (!(((unsigned char *)--b)[O] & M))
1544 {
1545 eio_dirent tmp = *a; *a = *b; *b = tmp;
1546 ++a;
1547 }
1548 while (b > a);
1549
1550 /* next bit, or stop, if no bits left in this path */
1551 if (!*--bit)
1552 break;
1553
1554 base_stk [stk_idx] = a;
1555 end_stk [stk_idx] = end;
1556 bit_stk [stk_idx] = bit;
1557 ++stk_idx;
1558
1559 end = a;
1560 }
1561 }
1562 while (stk_idx--);
1563 }
1564}
1565
1566static void
1567eio_dent_insertion_sort (eio_dirent *dents, int size)
1568{
1569 /* first move the smallest element to the front, to act as a sentinel */
1570 {
1571 int i;
1572 eio_dirent *min = dents;
1573
1574 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1575 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1576 if (EIO_DENT_CMP (dents [i], <, *min))
1577 min = &dents [i];
1578
1579 /* swap elements 0 and j (minimum) */
1580 {
1581 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1582 }
1583 }
1584
1585 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1586 {
1587 eio_dirent *i, *j;
1588
1589 for (i = dents + 1; i < dents + size; ++i)
1590 {
1591 eio_dirent value = *i;
1592
1593 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1594 j [1] = j [0];
1595
1596 j [1] = value;
1597 }
1598 }
1599}
1600
1601static void
1602eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1603{
1604 if (size <= 1)
1605 return; /* our insertion sort relies on size > 0 */
1606
1607 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1608 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1609 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1610
1611 /* use an insertion sort at the end, or for small arrays, */
1612 /* as insertion sort is more efficient for small partitions */
1613 eio_dent_insertion_sort (dents, size);
1614}
1615
718/* read a full directory */ 1616/* read a full directory */
719static void 1617static void
720eio__scandir (eio_req *req, worker *self) 1618eio__scandir (eio_req *req, etp_worker *self)
721{ 1619{
722 DIR *dirp; 1620 DIR *dirp;
723 union
724 {
725 EIO_STRUCT_DIRENT d;
726 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727 } *u;
728 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
729 char *name, *names; 1622 char *name, *names;
730 int memlen = 4096; 1623 int namesalloc = 4096;
731 int memofs = 0; 1624 int namesoffs = 0;
1625 int flags = req->int1;
1626 eio_dirent *dents = 0;
1627 int dentalloc = 128;
732 int res = 0; 1628 int dentoffs = 0;
1629 ino_t inode_bits = 0;
1630
1631 req->result = -1;
1632
1633 if (!(flags & EIO_READDIR_DENTS))
1634 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
733 1635
734 X_LOCK (wrklock); 1636 X_LOCK (wrklock);
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */
735 self->dirp = dirp = opendir (req->ptr1); 1638 self->dirp = dirp = opendir (req->ptr1);
736 self->dbuf = u = malloc (sizeof (*u)); 1639
737 req->flags |= EIO_FLAG_PTR2_FREE; 1640 if (req->flags & EIO_FLAG_PTR1_FREE)
1641 free (req->ptr1);
1642
1643 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1644 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
738 req->ptr2 = names = malloc (memlen); 1645 req->ptr2 = names = malloc (namesalloc);
739 X_UNLOCK (wrklock); 1646 X_UNLOCK (wrklock);
740 1647
741 if (dirp && u && names) 1648 if (dirp && names && (!flags || dents))
742 for (;;) 1649 for (;;)
743 { 1650 {
744 errno = 0; 1651 errno = 0;
745 readdir_r (dirp, &u->d, &entp); 1652 entp = readdir (dirp);
746 1653
747 if (!entp) 1654 if (!entp)
1655 {
1656 if (errno)
1657 break;
1658
1659 /* sort etc. */
1660 req->int1 = flags;
1661 req->result = dentoffs;
1662
1663 if (flags & EIO_READDIR_STAT_ORDER)
1664 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1665 else if (flags & EIO_READDIR_DIRS_FIRST)
1666 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1668 else
1669 {
1670 /* in this case, all is known, and we just put dirs first and sort them */
1671 eio_dirent *oth = dents + dentoffs;
1672 eio_dirent *dir = dents;
1673
1674 /* now partition dirs to the front, and non-dirs to the back */
1675 /* by walking from both sides and swapping if necessary */
1676 while (oth > dir)
1677 {
1678 if (dir->type == EIO_DT_DIR)
1679 ++dir;
1680 else if ((--oth)->type == EIO_DT_DIR)
1681 {
1682 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1683
1684 ++dir;
1685 }
1686 }
1687
1688 /* now sort the dirs only (dirs all have the same score) */
1689 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1690 }
1691
748 break; 1692 break;
1693 }
749 1694
1695 /* now add the entry to our list(s) */
750 name = entp->d_name; 1696 name = entp->d_name;
751 1697
1698 /* skip . and .. entries */
752 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753 { 1700 {
754 int len = strlen (name) + 1; 1701 int len = D_NAMLEN (entp) + 1;
755 1702
756 res++; 1703 while (ecb_expect_false (namesoffs + len > namesalloc))
757
758 while (memofs + len > memlen)
759 { 1704 {
760 memlen *= 2; 1705 namesalloc *= 2;
761 X_LOCK (wrklock); 1706 X_LOCK (wrklock);
762 req->ptr2 = names = realloc (names, memlen); 1707 req->ptr2 = names = realloc (names, namesalloc);
763 X_UNLOCK (wrklock); 1708 X_UNLOCK (wrklock);
764 1709
765 if (!names) 1710 if (!names)
766 break; 1711 break;
767 } 1712 }
768 1713
769 memcpy (names + memofs, name, len); 1714 memcpy (names + namesoffs, name, len);
1715
1716 if (dents)
1717 {
1718 struct eio_dirent *ent;
1719
1720 if (ecb_expect_false (dentoffs == dentalloc))
1721 {
1722 dentalloc *= 2;
1723 X_LOCK (wrklock);
1724 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725 X_UNLOCK (wrklock);
1726
1727 if (!dents)
1728 break;
1729 }
1730
1731 ent = dents + dentoffs;
1732
1733 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1734 ent->namelen = len - 1;
1735 ent->inode = D_INO (entp);
1736
1737 inode_bits |= ent->inode;
1738
1739 switch (D_TYPE (entp))
1740 {
1741 default:
1742 ent->type = EIO_DT_UNKNOWN;
1743 flags |= EIO_READDIR_FOUND_UNKNOWN;
1744 break;
1745
1746 #ifdef DT_FIFO
1747 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1748 #endif
1749 #ifdef DT_CHR
1750 case DT_CHR: ent->type = EIO_DT_CHR; break;
1751 #endif
1752 #ifdef DT_MPC
1753 case DT_MPC: ent->type = EIO_DT_MPC; break;
1754 #endif
1755 #ifdef DT_DIR
1756 case DT_DIR: ent->type = EIO_DT_DIR; break;
1757 #endif
1758 #ifdef DT_NAM
1759 case DT_NAM: ent->type = EIO_DT_NAM; break;
1760 #endif
1761 #ifdef DT_BLK
1762 case DT_BLK: ent->type = EIO_DT_BLK; break;
1763 #endif
1764 #ifdef DT_MPB
1765 case DT_MPB: ent->type = EIO_DT_MPB; break;
1766 #endif
1767 #ifdef DT_REG
1768 case DT_REG: ent->type = EIO_DT_REG; break;
1769 #endif
1770 #ifdef DT_NWK
1771 case DT_NWK: ent->type = EIO_DT_NWK; break;
1772 #endif
1773 #ifdef DT_CMP
1774 case DT_CMP: ent->type = EIO_DT_CMP; break;
1775 #endif
1776 #ifdef DT_LNK
1777 case DT_LNK: ent->type = EIO_DT_LNK; break;
1778 #endif
1779 #ifdef DT_SOCK
1780 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1781 #endif
1782 #ifdef DT_DOOR
1783 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1784 #endif
1785 #ifdef DT_WHT
1786 case DT_WHT: ent->type = EIO_DT_WHT; break;
1787 #endif
1788 }
1789
1790 ent->score = 7;
1791
1792 if (flags & EIO_READDIR_DIRS_FIRST)
1793 {
1794 if (ent->type == EIO_DT_UNKNOWN)
1795 {
1796 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1797 ent->score = 1;
1798 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1799 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1800 }
1801 else if (ent->type == EIO_DT_DIR)
1802 ent->score = 0;
1803 }
1804 }
1805
770 memofs += len; 1806 namesoffs += len;
1807 ++dentoffs;
1808 }
1809
1810 if (EIO_CANCELLED (req))
1811 {
1812 errno = ECANCELED;
1813 break;
771 } 1814 }
772 } 1815 }
773
774 if (errno)
775 res = -1;
776
777 req->result = res;
778} 1816}
779 1817
780/*****************************************************************************/ 1818/*****************************************************************************/
781 1819
782#define ALLOC(len) \ 1820#define ALLOC(len) \
… …
792 req->result = -1; \ 1830 req->result = -1; \
793 break; \ 1831 break; \
794 } \ 1832 } \
795 } 1833 }
796 1834
797X_THREAD_PROC (eio_proc) 1835X_THREAD_PROC (etp_proc)
798{ 1836{
799 eio_req *req; 1837 ETP_REQ *req;
800 struct timespec ts; 1838 struct timespec ts;
801 worker *self = (worker *)thr_arg; 1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
802 1841
803 /* try to distribute timeouts somewhat randomly */ 1842 /* try to distribute timeouts somewhat evenly */
804 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805 1844
806 for (;;) 1845 for (;;)
807 { 1846 {
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1847 ts.tv_sec = 0;
809 1848
810 X_LOCK (reqlock); 1849 X_LOCK (reqlock);
811 1850
812 for (;;) 1851 for (;;)
813 { 1852 {
814 self->req = req = reqq_shift (&req_queue); 1853 self->req = req = reqq_shift (&req_queue);
815 1854
816 if (req) 1855 if (req)
817 break; 1856 break;
818 1857
1858 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1859 {
1860 X_UNLOCK (reqlock);
1861 X_LOCK (wrklock);
1862 --started;
1863 X_UNLOCK (wrklock);
1864 goto quit;
1865 }
1866
819 ++idle; 1867 ++idle;
820 1868
821 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1869 if (idle <= max_idle)
1870 /* we are allowed to idle, so do so without any timeout */
1871 X_COND_WAIT (reqwait, reqlock);
1872 else
822 { 1873 {
823 if (idle > max_idle) 1874 /* initialise timeout once */
824 { 1875 if (!ts.tv_sec)
825 --idle; 1876 ts.tv_sec = time (0) + idle_timeout;
826 X_UNLOCK (reqlock);
827 X_LOCK (wrklock);
828 --started;
829 X_UNLOCK (wrklock);
830 goto quit;
831 }
832 1877
833 /* we are allowed to idle, so do so without any timeout */
834 X_COND_WAIT (reqwait, reqlock); 1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
835 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
836 } 1880 }
837 1881
838 --idle; 1882 --idle;
839 } 1883 }
840 1884
841 --nready; 1885 --nready;
842 1886
843 X_UNLOCK (reqlock); 1887 X_UNLOCK (reqlock);
844 1888
845 errno = 0; /* strictly unnecessary */
846
847 if (!EIO_CANCELLED (req))
848 switch (req->type) 1889 if (req->type < 0)
849 {
850 case EIO_READ: ALLOC (req->size);
851 req->result = req->offs >= 0
852 ? pread (req->int1, req->ptr2, req->size, req->offs)
853 : read (req->int1, req->ptr2, req->size); break;
854 case EIO_WRITE: req->result = req->offs >= 0
855 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
856 : write (req->int1, req->ptr2, req->size); break;
857
858 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876 case EIO_CLOSE: req->result = close (req->int1); break;
877 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886 case EIO_READLINK: ALLOC (NAME_MAX);
887 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889 case EIO_SYNC: req->result = 0; sync (); break;
890 case EIO_FSYNC: req->result = fsync (req->int1); break;
891 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893 case EIO_READDIR: eio__scandir (req, self); break;
894
895 case EIO_BUSY:
896#ifdef _WIN32
897 Sleep (req->nv1 * 1000.);
898#else
899 {
900 struct timeval tv;
901
902 tv.tv_sec = req->nv1;
903 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905 req->result = select (0, 0, 0, 0, &tv);
906 }
907#endif
908 break;
909
910 case EIO_UTIME:
911 case EIO_FUTIME:
912 {
913 struct timeval tv[2];
914 struct timeval *times;
915
916 if (req->nv1 != -1. || req->nv2 != -1.)
917 {
918 tv[0].tv_sec = req->nv1;
919 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920 tv[1].tv_sec = req->nv2;
921 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923 times = tv;
924 }
925 else
926 times = 0;
927
928
929 req->result = req->type == EIO_FUTIME
930 ? futimes (req->int1, times)
931 : utimes (req->ptr1, times);
932 }
933
934 case EIO_GROUP:
935 case EIO_NOP:
936 req->result = 0;
937 break;
938
939 case EIO_QUIT:
940 goto quit; 1890 goto quit;
941 1891
942 default: 1892 ETP_EXECUTE (self, req);
943 req->result = -1;
944 break;
945 }
946
947 req->errorno = errno;
948 1893
949 X_LOCK (reslock); 1894 X_LOCK (reslock);
950 1895
951 ++npending; 1896 ++npending;
952 1897
953 if (!reqq_push (&res_queue, req) && want_poll_cb) 1898 if (!reqq_push (&res_queue, req) && want_poll_cb)
954 want_poll_cb (); 1899 want_poll_cb ();
955 1900
956 self->req = 0; 1901 self->req = 0;
957 worker_clear (self); 1902 etp_worker_clear (self);
958 1903
959 X_UNLOCK (reslock); 1904 X_UNLOCK (reslock);
960 } 1905 }
961 1906
962quit: 1907quit:
963 X_LOCK (wrklock); 1908 X_LOCK (wrklock);
964 worker_free (self); 1909 etp_worker_free (self);
965 X_UNLOCK (wrklock); 1910 X_UNLOCK (wrklock);
966 1911
967 return 0; 1912 return 0;
968} 1913}
969 1914
970/*****************************************************************************/ 1915/*****************************************************************************/
971 1916
972static void eio_atfork_prepare (void) 1917int ecb_cold
973{
974 X_LOCK (wrklock);
975 X_LOCK (reqlock);
976 X_LOCK (reslock);
977#if !HAVE_PREADWRITE
978 X_LOCK (preadwritelock);
979#endif
980#if !HAVE_READDIR_R
981 X_LOCK (readdirlock);
982#endif
983}
984
985static void eio_atfork_parent (void)
986{
987#if !HAVE_READDIR_R
988 X_UNLOCK (readdirlock);
989#endif
990#if !HAVE_PREADWRITE
991 X_UNLOCK (preadwritelock);
992#endif
993 X_UNLOCK (reslock);
994 X_UNLOCK (reqlock);
995 X_UNLOCK (wrklock);
996}
997
998static void eio_atfork_child (void)
999{
1000 eio_req *prv;
1001
1002 while (prv = reqq_shift (&req_queue))
1003 eio_destroy (prv);
1004
1005 while (prv = reqq_shift (&res_queue))
1006 eio_destroy (prv);
1007
1008 while (wrk_first.next != &wrk_first)
1009 {
1010 worker *wrk = wrk_first.next;
1011
1012 if (wrk->req)
1013 eio_destroy (wrk->req);
1014
1015 worker_clear (wrk);
1016 worker_free (wrk);
1017 }
1018
1019 started = 0;
1020 idle = 0;
1021 nreqs = 0;
1022 nready = 0;
1023 npending = 0;
1024
1025 eio_atfork_parent ();
1026}
1027
1028int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1918eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029{ 1919{
1030 want_poll_cb = want_poll; 1920 return etp_init (want_poll, done_poll);
1031 done_poll_cb = done_poll;
1032
1033#ifdef _WIN32
1034 X_MUTEX_CHECK (wrklock);
1035 X_MUTEX_CHECK (reslock);
1036 X_MUTEX_CHECK (reqlock);
1037 X_MUTEX_CHECK (reqwait);
1038 X_MUTEX_CHECK (preadwritelock);
1039 X_MUTEX_CHECK (readdirlock);
1040
1041 X_COND_CHECK (reqwait);
1042#endif
1043
1044 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045} 1921}
1046 1922
1923ecb_inline void
1047static void eio_api_destroy (eio_req *req) 1924eio_api_destroy (eio_req *req)
1048{ 1925{
1049 free (req); 1926 free (req);
1050} 1927}
1051 1928
1052#define REQ(rtype) \ 1929#define REQ(rtype) \
… …
1071 { \ 1948 { \
1072 eio_api_destroy (req); \ 1949 eio_api_destroy (req); \
1073 return 0; \ 1950 return 0; \
1074 } 1951 }
1075 1952
1953static void
1954eio_execute (etp_worker *self, eio_req *req)
1955{
1956 if (ecb_expect_false (EIO_CANCELLED (req)))
1957 {
1958 req->result = -1;
1959 req->errorno = ECANCELED;
1960 return;
1961 }
1962
1963 switch (req->type)
1964 {
1965 case EIO_READ: ALLOC (req->size);
1966 req->result = req->offs >= 0
1967 ? pread (req->int1, req->ptr2, req->size, req->offs)
1968 : read (req->int1, req->ptr2, req->size); break;
1969 case EIO_WRITE: req->result = req->offs >= 0
1970 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1971 : write (req->int1, req->ptr2, req->size); break;
1972
1973 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1974 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1975
1976 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1977 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1978 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1979 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1980 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1981 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1982
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1987
1988 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1989 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1990 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1991 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1992 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1993 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1994
1995 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1996 case EIO_CLOSE: req->result = close (req->int1); break;
1997 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1998 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1999 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2000 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2001 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2002 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2003 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2004 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2005
2006 case EIO_REALPATH: eio__realpath (req, self); break;
2007
2008 case EIO_READLINK: ALLOC (PATH_MAX);
2009 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2010
2011 case EIO_SYNC: req->result = 0; sync (); break;
2012 case EIO_FSYNC: req->result = fsync (req->int1); break;
2013 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2014 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2015 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2017 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2020
2021 case EIO_READDIR: eio__scandir (req, self); break;
2022
2023 case EIO_BUSY:
2024#ifdef _WIN32
2025 Sleep (req->nv1 * 1e3);
2026#else
2027 {
2028 struct timeval tv;
2029
2030 tv.tv_sec = req->nv1;
2031 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2032
2033 req->result = select (0, 0, 0, 0, &tv);
2034 }
2035#endif
2036 break;
2037
2038 case EIO_UTIME:
2039 case EIO_FUTIME:
2040 {
2041 struct timeval tv[2];
2042 struct timeval *times;
2043
2044 if (req->nv1 != -1. || req->nv2 != -1.)
2045 {
2046 tv[0].tv_sec = req->nv1;
2047 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2048 tv[1].tv_sec = req->nv2;
2049 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2050
2051 times = tv;
2052 }
2053 else
2054 times = 0;
2055
2056 req->result = req->type == EIO_FUTIME
2057 ? futimes (req->int1, times)
2058 : utimes (req->ptr1, times);
2059 }
2060 break;
2061
2062 case EIO_GROUP:
2063 abort (); /* handled in eio_request */
2064
2065 case EIO_NOP:
2066 req->result = 0;
2067 break;
2068
2069 case EIO_CUSTOM:
2070 req->feed (req);
2071 break;
2072
2073 default:
2074 errno = ENOSYS;
2075 req->result = -1;
2076 break;
2077 }
2078
2079 req->errorno = errno;
2080}
2081
1076#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
1077 2083
1078eio_req *eio_nop (int pri, eio_cb cb, void *data) 2084eio_req *eio_nop (int pri, eio_cb cb, void *data)
1079{ 2085{
1080 REQ (EIO_NOP); SEND; 2086 REQ (EIO_NOP); SEND;
… …
1093eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2099eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1094{ 2100{
1095 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2101 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1096} 2102}
1097 2103
2104eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2105{
2106 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2107}
2108
2109eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2110{
2111 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2112}
2113
2114eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2115{
2116 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2117}
2118
2119eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122}
2123
2124eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127}
2128
2129eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132}
2133
1098eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1099{ 2135{
1100 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1101} 2137}
1102 2138
… …
1123eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1124{ 2160{
1125 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2161 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1126} 2162}
1127 2163
2164eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2167}
2168
1128eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2169eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1129{ 2170{
1130 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2171 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1131} 2172}
1132 2173
… …
1138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2179eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1139{ 2180{
1140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2181 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1141} 2182}
1142 2183
1143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2184eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1144{ 2185{
1145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2186 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1146} 2187}
1147 2188
1148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2189eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2209eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1169{ 2210{
1170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2211 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1171} 2212}
1172 2213
1173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1174{ 2215{
1175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1176} 2217}
1177 2218
1178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2219eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2235eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1195{ 2236{
1196 return eio__1path (EIO_READLINK, path, pri, cb, data); 2237 return eio__1path (EIO_READLINK, path, pri, cb, data);
1197} 2238}
1198 2239
2240eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2241{
2242 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2243}
2244
1199eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2245eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1200{ 2246{
1201 return eio__1path (EIO_STAT, path, pri, cb, data); 2247 return eio__1path (EIO_STAT, path, pri, cb, data);
1202} 2248}
1203 2249
1204eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2250eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1205{ 2251{
1206 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2252 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1207} 2253}
1208 2254
2255eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2256{
2257 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2258}
2259
1209eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2260eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1210{ 2261{
1211 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2262 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1212} 2263}
1213 2264
1214eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2265eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1215{ 2266{
1216 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2267 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1217} 2268}
1218 2269
1219eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2270eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1220{ 2271{
1221 return eio__1path (EIO_READDIR, path, pri, cb, data); 2272 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1222} 2273}
1223 2274
1224eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2275eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1225{ 2276{
1226 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2277 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1227} 2278}
1228 2279
1229static eio_req * 2280static eio_req *
1230eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2281eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1231{ 2282{
… …
1255eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2306eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1256{ 2307{
1257 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2308 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1258} 2309}
1259 2310
2311eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2312{
2313 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2314}
2315
1260#endif 2316#endif
1261 2317
1262eio_req *eio_grp (eio_cb cb, void *data) 2318eio_req *eio_grp (eio_cb cb, void *data)
1263{ 2319{
1264 const int pri = EIO_PRI_MAX; 2320 const int pri = EIO_PRI_MAX;
… …
1271#undef SEND 2327#undef SEND
1272 2328
1273/*****************************************************************************/ 2329/*****************************************************************************/
1274/* grp functions */ 2330/* grp functions */
1275 2331
2332void
1276void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2333eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1277{ 2334{
1278 grp->int2 = limit; 2335 grp->int2 = limit;
1279 grp->feed = feed; 2336 grp->feed = feed;
1280 2337
1281 grp_try_feed (grp); 2338 grp_try_feed (grp);
1282} 2339}
1283 2340
2341void
1284void eio_grp_limit (eio_req *grp, int limit) 2342eio_grp_limit (eio_req *grp, int limit)
1285{ 2343{
1286 grp->int2 = limit; 2344 grp->int2 = limit;
1287 2345
1288 grp_try_feed (grp); 2346 grp_try_feed (grp);
1289} 2347}
1290 2348
2349void
1291void eio_grp_add (eio_req *grp, eio_req *req) 2350eio_grp_add (eio_req *grp, eio_req *req)
1292{ 2351{
1293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2352 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2353
2354 grp->flags |= EIO_FLAG_GROUPADD;
1294 2355
1295 ++grp->size; 2356 ++grp->size;
1296 req->grp = grp; 2357 req->grp = grp;
1297 2358
1298 req->grp_prev = 0; 2359 req->grp_prev = 0;
… …
1305} 2366}
1306 2367
1307/*****************************************************************************/ 2368/*****************************************************************************/
1308/* misc garbage */ 2369/* misc garbage */
1309 2370
2371eio_ssize_t
1310ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1311{ 2373{
1312 worker wrk; 2374 etp_worker wrk;
2375 eio_ssize_t ret;
1313 2376
1314 wrk.dbuf = 0; 2377 wrk.dbuf = 0;
1315 2378
1316 eio__sendfile (ofd, ifd, offset, count, &wrk); 2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1317 2380
1318 if (wrk.dbuf) 2381 if (wrk.dbuf)
1319 free (wrk.dbuf); 2382 free (wrk.dbuf);
1320}
1321 2383
2384 return ret;
2385}
2386

Diff Legend

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