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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.89 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
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72
73static void eio_destroy (eio_req *req);
18 74
19#ifndef EIO_FINISH 75#ifndef EIO_FINISH
20# 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
21#endif 77#endif
22 78
26 82
27#ifndef EIO_FEED 83#ifndef EIO_FEED
28# 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)
29#endif 85#endif
30 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
31#ifdef _WIN32 98#ifdef _WIN32
32 99
33 /*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 }
34 130
35#else 131#else
36 132
37# include "config.h"
38# include <sys/time.h> 133 #include <sys/time.h>
39# include <sys/select.h> 134 #include <sys/select.h>
135 #include <sys/statvfs.h>
40# include <unistd.h> 136 #include <unistd.h>
41# include <utime.h> 137 #include <utime.h>
42# include <signal.h> 138 #include <signal.h>
139 #include <dirent.h>
43 140
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
44# ifndef EIO_STRUCT_DIRENT 163 #ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 164 #define EIO_STRUCT_DIRENT struct dirent
165 #endif
166
46# endif 167#endif
47
48#endif
49
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53 168
54#if HAVE_SENDFILE 169#if HAVE_SENDFILE
55# if __linux 170# if __linux
56# include <sys/sendfile.h> 171# include <sys/sendfile.h>
57# elif __freebsd 172# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 173# include <sys/socket.h>
59# include <sys/uio.h> 174# include <sys/uio.h>
60# elif __hpux 175# elif __hpux
61# include <sys/socket.h> 176# include <sys/socket.h>
62# elif __solaris /* not yet */ 177# elif __solaris
63# include <sys/sendfile.h> 178# include <sys/sendfile.h>
64# else 179# else
65# error sendfile support requested but not available 180# error sendfile support requested but not available
66# endif 181# endif
67#endif 182#endif
68 183
69/* number of seconds after which an idle threads exit */ 184#ifndef D_TYPE
70#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
71 193
72/* used for struct dirent, AIX doesn't provide it */ 194/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 195#ifndef NAME_MAX
74# define NAME_MAX 4096 196# define NAME_MAX 4096
75#endif 197#endif
76 198
199/* used for readlink etc. */
200#ifndef PATH_MAX
201# define PATH_MAX 4096
202#endif
203
77/* buffer size for various temporary buffers */ 204/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 205#define EIO_BUFSIZE 65536
79 206
80#define dBUF \ 207#define dBUF \
81 char *eio_buf; \ 208 char *eio_buf; \
82 X_LOCK (wrklock); \ 209 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 210 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 211 ETP_WORKER_UNLOCK (self); \
212 errno = ENOMEM; \
85 if (!eio_buf) \ 213 if (!eio_buf) \
86 return -1; 214 return -1;
87 215
88#define EIO_TICKS ((1000000 + 1023) >> 10) 216#define EIO_TICKS ((1000000 + 1023) >> 10)
89 217
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
90static void (*want_poll_cb) (void); 261static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 262static void (*done_poll_cb) (void);
92 263
93static unsigned int max_poll_time = 0; 264static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 265static unsigned int max_poll_reqs; /* reslock */
95 266
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 267static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 268static volatile unsigned int nready; /* reqlock */
98{ 269static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 272
103static unsigned int started, idle, wanted; 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)
104 304
105/* worker threads management */ 305/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 306
108typedef struct worker { 307static void ecb_cold
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 308etp_worker_clear (etp_worker *wrk)
123{ 309{
124 if (wrk->dirp) 310 ETP_WORKER_CLEAR (wrk);
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 311}
136 312
137static void worker_free (worker *wrk) 313static void ecb_cold
314etp_worker_free (etp_worker *wrk)
138{ 315{
139 wrk->next->prev = wrk->prev; 316 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 317 wrk->prev->next = wrk->next;
141 318
142 free (wrk); 319 free (wrk);
143} 320}
144 321
145static volatile unsigned int nreqs, nready, npending; 322static unsigned int
146static volatile unsigned int max_idle = 4; 323etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 324{
325 int retval;
326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
327 retval = nreqs;
328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 329 return retval;
155} 330}
156 331
157unsigned int eio_nready (void) 332static unsigned int
333etp_nready (void)
158{ 334{
159 unsigned int retval; 335 unsigned int retval;
160 336
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 338 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 340
165 return retval; 341 return retval;
166} 342}
167 343
168unsigned int eio_npending (void) 344static unsigned int
345etp_npending (void)
169{ 346{
170 unsigned int retval; 347 unsigned int retval;
171 348
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 350 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 352
176 return retval; 353 return retval;
177} 354}
178 355
179unsigned int eio_nthreads (void) 356static unsigned int
357etp_nthreads (void)
180{ 358{
181 unsigned int retval; 359 unsigned int retval;
182 360
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 362 retval = started;
191 * a somewhat faster data structure might be nice, but 369 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 370 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 371 * per shift, the most expensive operation.
194 */ 372 */
195typedef struct { 373typedef struct {
196 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 */
197 int size; 375 int size;
198} reqq; 376} etp_reqq;
199 377
200static reqq req_queue; 378static etp_reqq req_queue;
201static reqq res_queue; 379static etp_reqq res_queue;
202 380
203static int reqq_push (reqq *q, eio_req *req) 381static int ecb_noinline
382reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 383{
205 int pri = req->pri; 384 int pri = req->pri;
206 req->next = 0; 385 req->next = 0;
207 386
208 if (q->qe[pri]) 387 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 393 q->qe[pri] = q->qs[pri] = req;
215 394
216 return q->size++; 395 return q->size++;
217} 396}
218 397
219static eio_req *reqq_shift (reqq *q) 398static ETP_REQ * ecb_noinline
399reqq_shift (etp_reqq *q)
220{ 400{
221 int pri; 401 int pri;
222 402
223 if (!q->size) 403 if (!q->size)
224 return 0; 404 return 0;
225 405
226 --q->size; 406 --q->size;
227 407
228 for (pri = EIO_NUM_PRI; pri--; ) 408 for (pri = ETP_NUM_PRI; pri--; )
229 { 409 {
230 eio_req *req = q->qs[pri]; 410 eio_req *req = q->qs[pri];
231 411
232 if (req) 412 if (req)
233 { 413 {
239 } 419 }
240 420
241 abort (); 421 abort ();
242} 422}
243 423
244static void grp_try_feed (eio_req *grp) 424static void ecb_cold
425etp_thread_init (void)
245{ 426{
246 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)
247 { 458 {
248 int old_len = grp->size; 459 etp_worker *wrk = wrk_first.next;
249 460
250 EIO_FEED (grp); 461 if (wrk->req)
462 ETP_DESTROY (wrk->req);
251 463
252 /* stop if no progress has been made */ 464 etp_worker_clear (wrk);
253 if (old_len == grp->size) 465 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 466 }
259}
260 467
261static int eio_finish (eio_req *req); 468 started = 0;
469 idle = 0;
470 nreqs = 0;
471 nready = 0;
472 npending = 0;
262 473
263static int grp_dec (eio_req *grp) 474 etp_thread_init ();
264{ 475}
265 --grp->size;
266 476
267 /* call feeder, if applicable */ 477static void ecb_cold
268 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}
269 483
270 /* finish, if done */ 484static int ecb_cold
271 if (!grp->size && grp->int1) 485etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 486{
273 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
274 return 0; 494 return 0;
275} 495}
276 496
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 497X_THREAD_PROC (etp_proc);
326 498
327static void start_thread (void) 499static void ecb_cold
500etp_start_thread (void)
328{ 501{
329 worker *wrk = calloc (1, sizeof (worker)); 502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 503
331 if (!wrk) 504 /*TODO*/
332 croak ("unable to allocate worker thread data"); 505 assert (("unable to allocate worker thread data", wrk));
333 506
334 X_LOCK (wrklock); 507 X_LOCK (wrklock);
335 508
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 509 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 510 {
338 wrk->prev = &wrk_first; 511 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 512 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 513 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 514 wrk_first.next = wrk;
345 free (wrk); 518 free (wrk);
346 519
347 X_UNLOCK (wrklock); 520 X_UNLOCK (wrklock);
348} 521}
349 522
523static void
350static void maybe_start_thread (void) 524etp_maybe_start_thread (void)
351{ 525{
352 if (eio_nthreads () >= wanted) 526 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 527 return;
354 528
355 /* todo: maybe use idle here, but might be less exact */ 529 /* todo: maybe use idle here, but might be less exact */
356 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 ()))
357 return; 531 return;
358 532
359 start_thread (); 533 etp_start_thread ();
360} 534}
361 535
362void eio_submit (eio_req *req) 536static void ecb_cold
363{ 537etp_end_thread (void)
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{ 538{
377 eio_req *req = calloc (1, sizeof (eio_req)); 539 eio_req *req = calloc (1, sizeof (eio_req));
378 540
379 req->type = EIO_QUIT; 541 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 543
382 X_LOCK (reqlock); 544 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 545 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 546 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 547 X_UNLOCK (reqlock);
387 X_LOCK (wrklock); 549 X_LOCK (wrklock);
388 --started; 550 --started;
389 X_UNLOCK (wrklock); 551 X_UNLOCK (wrklock);
390} 552}
391 553
392void eio_set_max_poll_time (double nseconds) 554static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 555etp_poll (void)
429{ 556{
430 int maxreqs = max_poll_reqs; 557 unsigned int maxreqs;
558 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 559 struct timeval tv_start, tv_now;
432 eio_req *req;
433 560
561 X_LOCK (reslock);
562 maxreqs = max_poll_reqs;
563 maxtime = max_poll_time;
564 X_UNLOCK (reslock);
565
434 if (max_poll_time) 566 if (maxtime)
435 gettimeofday (&tv_start, 0); 567 gettimeofday (&tv_start, 0);
436 568
437 for (;;) 569 for (;;)
438 { 570 {
571 ETP_REQ *req;
572
439 maybe_start_thread (); 573 etp_maybe_start_thread ();
440 574
441 X_LOCK (reslock); 575 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 576 req = reqq_shift (&res_queue);
443 577
444 if (req) 578 if (req)
452 X_UNLOCK (reslock); 586 X_UNLOCK (reslock);
453 587
454 if (!req) 588 if (!req)
455 return 0; 589 return 0;
456 590
591 X_LOCK (reqlock);
457 --nreqs; 592 --nreqs;
593 X_UNLOCK (reqlock);
458 594
459 if (req->type == EIO_GROUP && req->size) 595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 596 {
461 req->int1 = 1; /* mark request as delayed */ 597 req->int1 = 1; /* mark request as delayed */
462 continue; 598 continue;
463 } 599 }
464 else 600 else
465 { 601 {
466 int res = eio_finish (req); 602 int res = ETP_FINISH (req);
467 if (res) 603 if (ecb_expect_false (res))
468 return res; 604 return res;
469 } 605 }
470 606
471 if (maxreqs && !--maxreqs) 607 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 608 break;
473 609
474 if (max_poll_time) 610 if (maxtime)
475 { 611 {
476 gettimeofday (&tv_now, 0); 612 gettimeofday (&tv_now, 0);
477 613
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 614 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 615 break;
480 } 616 }
481 } 617 }
482 618
483 errno = EAGAIN; 619 errno = EAGAIN;
484 return -1; 620 return -1;
485} 621}
486 622
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
487/*****************************************************************************/ 873/*****************************************************************************/
488/* work around various missing functions */ 874/* work around various missing functions */
489 875
490#if !HAVE_PREADWRITE 876#if !HAVE_PREADWRITE
877# undef pread
878# undef pwrite
491# define pread aio_pread 879# define pread eio__pread
492# define pwrite aio_pwrite 880# define pwrite eio__pwrite
493 881
494/* 882static eio_ssize_t
495 * make our pread/pwrite safe against themselves, but not against 883eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 884{
503 ssize_t res; 885 eio_ssize_t res;
504 off_t ooffset; 886 off_t ooffset;
505 887
506 X_LOCK (preadwritelock); 888 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR); 889 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET); 890 lseek (fd, offset, SEEK_SET);
511 X_UNLOCK (preadwritelock); 893 X_UNLOCK (preadwritelock);
512 894
513 return res; 895 return res;
514} 896}
515 897
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 898static eio_ssize_t
899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 900{
518 ssize_t res; 901 eio_ssize_t res;
519 off_t ooffset; 902 off_t ooffset;
520 903
521 X_LOCK (preadwritelock); 904 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 905 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 906 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 907 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 908 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 909 X_UNLOCK (preadwritelock);
527 910
528 return res; 911 return res;
529} 912}
530#endif 913#endif
531 914
532#ifndef HAVE_FUTIMES 915#ifndef HAVE_UTIMES
533 916
917# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 918# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 919
920static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 921eio__utimes (const char *filename, const struct timeval times[2])
538{ 922{
539 if (times) 923 if (times)
540 { 924 {
541 struct utimbuf buf; 925 struct utimbuf buf;
542 926
547 } 931 }
548 else 932 else
549 return utime (filename, 0); 933 return utime (filename, 0);
550} 934}
551 935
936#endif
937
938#ifndef HAVE_FUTIMES
939
940# undef futimes
941# define futimes(fd,times) eio__futimes (fd, times)
942
943static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 944eio__futimes (int fd, const struct timeval tv[2])
553{ 945{
554 errno = ENOSYS; 946 errno = ENOSYS;
555 return -1; 947 return -1;
556} 948}
557 949
558#endif 950#endif
559 951
560#if !HAVE_FDATASYNC 952#if !HAVE_FDATASYNC
953# undef fdatasync
561# define fdatasync fsync 954# define fdatasync(fd) fsync (fd)
562#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}
563 995
564#if !HAVE_READAHEAD 996#if !HAVE_READAHEAD
997# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 998# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 999
1000static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1001eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 1002{
569 size_t todo = count; 1003 size_t todo = count;
570 dBUF; 1004 dBUF;
571 1005
572 while (todo > 0) 1006 while (todo > 0)
573 { 1007 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1008 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 1009
576 pread (fd, aio_buf, len, offset); 1010 pread (fd, eio_buf, len, offset);
577 offset += len; 1011 offset += len;
578 todo -= len; 1012 todo -= len;
579 } 1013 }
580 1014
581 errno = 0; 1015 errno = 0;
582 return count; 1016 return count;
583} 1017}
584 1018
585#endif 1019#endif
586 1020
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 1021/* sendfile always needs emulation */
1022static eio_ssize_t
618static ssize_t 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)
619{ 1024{
1025 eio_ssize_t written = 0;
620 ssize_t res; 1026 eio_ssize_t res;
621 1027
622 if (!count) 1028 if (!count)
623 return 0; 1029 return 0;
624 1030
1031 for (;;)
1032 {
1033#ifdef __APPLE__
1034# undef HAVE_SENDFILE /* broken, as everything on os x */
1035#endif
625#if HAVE_SENDFILE 1036#if HAVE_SENDFILE
626# if __linux 1037# if __linux
1038 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1039 res = sendfile (ofd, ifd, &soffset, count);
628 1040
629# elif __freebsd 1041# elif __FreeBSD__
630 /* 1042 /*
631 * 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
632 * wasted on making it similar to other I/O functions. 1044 * wasted on making it similar to other I/O functions.
633 */ 1045 */
634 {
635 off_t sbytes; 1046 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1047 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1048
638 if (res < 0 && sbytes) 1049 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* 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)
640 res = sbytes; 1054 res = sbytes;
641 } 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;
642 1068
643# elif __hpux 1069# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1070 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1071
646# elif __solaris 1072# elif __solaris
647 {
648 struct sendfilevec vec; 1073 struct sendfilevec vec;
649 size_t sbytes; 1074 size_t sbytes;
650 1075
651 vec.sfv_fd = ifd; 1076 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1077 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1078 vec.sfv_off = offset;
654 vec.sfv_len = count; 1079 vec.sfv_len = count;
655 1080
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1081 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1082
658 if (res < 0 && sbytes) 1083 if (res < 0 && sbytes)
659 res = sbytes; 1084 res = sbytes;
660 }
661 1085
662# 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
663#else 1097#else
664 res = -1; 1098 res = -1;
665 errno = ENOSYS; 1099 errno = ENOSYS;
666#endif 1100#endif
667 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
668 if (res < 0 1125 if (res < 0
669 && (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 */
670#if __solaris 1132#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1134#endif
673 ) 1135 )
674 ) 1136 )
678 1140
679 res = 0; 1141 res = 0;
680 1142
681 while (count) 1143 while (count)
682 { 1144 {
683 ssize_t cnt; 1145 eio_ssize_t cnt;
684 1146
685 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);
686 1148
687 if (cnt <= 0) 1149 if (cnt <= 0)
688 { 1150 {
705 } 1167 }
706 1168
707 return res; 1169 return res;
708} 1170}
709 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
710/* read a full directory */ 1616/* read a full directory */
1617static void
711static void scandir_ (eio_req *req, worker *self) 1618eio__scandir (eio_req *req, etp_worker *self)
712{ 1619{
713 DIR *dirp; 1620 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1622 char *name, *names;
721 int memlen = 4096; 1623 int namesalloc = 4096;
722 int memofs = 0; 1624 int namesoffs = 0;
1625 int flags = req->int1;
1626 eio_dirent *dents = 0;
1627 int dentalloc = 128;
723 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);
724 1635
725 X_LOCK (wrklock); 1636 X_LOCK (wrklock);
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1638 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1639
728 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;
729 req->ptr2 = names = malloc (memlen); 1645 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1646 X_UNLOCK (wrklock);
731 1647
732 if (dirp && u && names) 1648 if (dirp && names && (!flags || dents))
733 for (;;) 1649 for (;;)
734 { 1650 {
735 errno = 0; 1651 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1652 entp = readdir (dirp);
737 1653
738 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
739 break; 1692 break;
1693 }
740 1694
1695 /* now add the entry to our list(s) */
741 name = entp->d_name; 1696 name = entp->d_name;
742 1697
1698 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1700 {
745 int len = strlen (name) + 1; 1701 int len = D_NAMLEN (entp) + 1;
746 1702
747 res++; 1703 while (ecb_expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1704 {
751 memlen *= 2; 1705 namesalloc *= 2;
752 X_LOCK (wrklock); 1706 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1707 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1708 X_UNLOCK (wrklock);
755 1709
756 if (!names) 1710 if (!names)
757 break; 1711 break;
758 } 1712 }
759 1713
760 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
761 memofs += len; 1806 namesoffs += len;
1807 ++dentoffs;
1808 }
1809
1810 if (EIO_CANCELLED (req))
1811 {
1812 errno = ECANCELED;
1813 break;
762 } 1814 }
763 } 1815 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769} 1816}
770 1817
771/*****************************************************************************/ 1818/*****************************************************************************/
772 1819
1820#define ALLOC(len) \
1821 if (!req->ptr2) \
1822 { \
1823 X_LOCK (wrklock); \
1824 req->flags |= EIO_FLAG_PTR2_FREE; \
1825 X_UNLOCK (wrklock); \
1826 req->ptr2 = malloc (len); \
1827 if (!req->ptr2) \
1828 { \
1829 errno = ENOMEM; \
1830 req->result = -1; \
1831 break; \
1832 } \
1833 }
1834
773X_THREAD_PROC (eio_proc) 1835X_THREAD_PROC (etp_proc)
774{ 1836{
775 eio_req *req; 1837 ETP_REQ *req;
776 struct timespec ts; 1838 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
778 1841
779 /* try to distribute timeouts somewhat randomly */ 1842 /* try to distribute timeouts somewhat evenly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1844
782 for (;;) 1845 for (;;)
783 { 1846 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1847 ts.tv_sec = 0;
785 1848
786 X_LOCK (reqlock); 1849 X_LOCK (reqlock);
787 1850
788 for (;;) 1851 for (;;)
789 { 1852 {
790 self->req = req = reqq_shift (&req_queue); 1853 self->req = req = reqq_shift (&req_queue);
791 1854
792 if (req) 1855 if (req)
793 break; 1856 break;
794 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
795 ++idle; 1867 ++idle;
796 1868
797 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
798 { 1873 {
799 if (idle > max_idle) 1874 /* initialise timeout once */
800 { 1875 if (!ts.tv_sec)
801 --idle; 1876 ts.tv_sec = time (0) + idle_timeout;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808 1877
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
811 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
812 } 1880 }
813 1881
814 --idle; 1882 --idle;
815 } 1883 }
816 1884
817 --nready; 1885 --nready;
818 1886
819 X_UNLOCK (reqlock); 1887 X_UNLOCK (reqlock);
820 1888
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type) 1889 if (req->type < 0)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit; 1890 goto quit;
910 1891
911 default: 1892 ETP_EXECUTE (self, req);
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1893
918 X_LOCK (reslock); 1894 X_LOCK (reslock);
919 1895
920 ++npending; 1896 ++npending;
921 1897
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1898 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1899 want_poll_cb ();
924 1900
925 self->req = 0; 1901 self->req = 0;
926 worker_clear (self); 1902 etp_worker_clear (self);
927 1903
928 X_UNLOCK (reslock); 1904 X_UNLOCK (reslock);
929 } 1905 }
930 1906
931quit: 1907quit:
932 X_LOCK (wrklock); 1908 X_LOCK (wrklock);
933 worker_free (self); 1909 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1910 X_UNLOCK (wrklock);
935 1911
936 return 0; 1912 return 0;
937} 1913}
938 1914
939/*****************************************************************************/ 1915/*****************************************************************************/
940 1916
941static void eio_atfork_prepare (void) 1917int ecb_cold
1918eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1919{
943 X_LOCK (wrklock); 1920 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1921}
953 1922
954static void eio_atfork_parent (void) 1923ecb_inline void
1924eio_api_destroy (eio_req *req)
955{ 1925{
956#if !HAVE_READDIR_R 1926 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1927}
966 1928
967static void eio_atfork_child (void) 1929#define REQ(rtype) \
968{ 1930 eio_req *req; \
969 eio_req *prv; 1931 \
1932 req = (eio_req *)calloc (1, sizeof *req); \
1933 if (!req) \
1934 return 0; \
1935 \
1936 req->type = rtype; \
1937 req->pri = pri; \
1938 req->finish = cb; \
1939 req->data = data; \
1940 req->destroy = eio_api_destroy;
970 1941
971 while (prv = reqq_shift (&req_queue)) 1942#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1943
974 while (prv = reqq_shift (&res_queue)) 1944#define PATH \
975 eio_destroy (prv); 1945 req->flags |= EIO_FLAG_PTR1_FREE; \
1946 req->ptr1 = strdup (path); \
1947 if (!req->ptr1) \
1948 { \
1949 eio_api_destroy (req); \
1950 return 0; \
1951 }
976 1952
977 while (wrk_first.next != &wrk_first) 1953static void
1954eio_execute (etp_worker *self, eio_req *req)
1955{
1956 if (ecb_expect_false (EIO_CANCELLED (req)))
978 { 1957 {
979 worker *wrk = wrk_first.next; 1958 req->result = -1;
1959 req->errorno = ECANCELED;
1960 return;
1961 }
980 1962
981 if (wrk->req) 1963 switch (req->type)
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 } 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;
987 1972
988 started = 0; 1973 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
989 idle = 0; 1974 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
990 nreqs = 0;
991 nready = 0;
992 npending = 0;
993 1975
994 eio_atfork_parent (); 1976 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
995} 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;
996 1982
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
998{ 1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
999 want_poll_cb = want_poll; 1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1000 done_poll_cb = done_poll; 1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1001 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:
1002#ifdef _WIN32 2024#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 2025 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 2026#else
1005 X_MUTEX_CHECK (reqlock); 2027 {
1006 X_MUTEX_CHECK (reqwait); 2028 struct timeval tv;
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009 2029
1010 X_COND_CHECK (reqwait); 2030 tv.tv_sec = req->nv1;
1011#endif 2031 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012 2032
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2033 req->result = select (0, 0, 0, 0, &tv);
1014} 2034 }
2035#endif
2036 break;
1015 2037
1016#if 0 2038 case EIO_UTIME:
2039 case EIO_FUTIME:
2040 {
2041 struct timeval tv[2];
2042 struct timeval *times;
1017 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
2082#ifndef EIO_NO_WRAPPERS
2083
2084eio_req *eio_nop (int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_NOP); SEND;
2087}
2088
2089eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2092}
2093
1018eio_req *eio_fsync (int fd, eio_cb cb); 2094eio_req *eio_sync (int pri, eio_cb cb, void *data)
2095{
2096 REQ (EIO_SYNC); SEND;
2097}
2098
2099eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2100{
2101 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2102}
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
1019eio_req *eio_fdatasync (int fd, eio_cb cb); 2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020eio_req *eio_dupclose (int fd, eio_cb cb); 2135{
2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2137}
2138
2139eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2140{
2141 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2142}
2143
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2144eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2145{
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2146 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2147}
2148
2149eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2152}
2153
2154eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2155{
2156 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2157}
2158
2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2160{
2161 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2162}
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
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2169eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2170{
2171 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2172}
2173
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2174eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2175{
2176 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2177}
2178
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2179eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2182}
2183
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2184eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2187}
2188
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2189eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2190{
2191 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2192}
2193
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2194eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2195{
2196 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2197}
2198
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2199eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2200{
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2201 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2202}
2203
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2204eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2205{
2206 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2207}
2208
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2209eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2210{
2211 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2212}
2213
2214eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2215{
2216 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2217}
2218
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2219eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2220{
2221 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2222}
2223
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2224eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2225{
2226 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2227}
2228
2229static eio_req *
2230eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2231{
2232 REQ (type); PATH; SEND;
2233}
2234
2235eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2236{
2237 return eio__1path (EIO_READLINK, path, pri, cb, data);
2238}
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
2245eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2246{
2247 return eio__1path (EIO_STAT, path, pri, cb, data);
2248}
2249
2250eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2251{
2252 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2253}
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
1040eio_req *eio_unlink (const char *path, eio_cb cb); 2260eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2261{
2262 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2263}
2264
1041eio_req *eio_rmdir (const char *path, eio_cb cb); 2265eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2266{
2267 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2268}
2269
2270eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2271{
2272 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2273}
2274
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2275eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2276{
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2277 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2278}
2279
2280static eio_req *
2281eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2282{
2283 REQ (type); PATH;
2284
2285 req->flags |= EIO_FLAG_PTR2_FREE;
2286 req->ptr2 = strdup (new_path);
2287 if (!req->ptr2)
2288 {
2289 eio_api_destroy (req);
2290 return 0;
2291 }
2292
2293 SEND;
2294}
2295
2296eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2297{
2298 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2299}
2300
2301eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2302{
2303 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2304}
2305
2306eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2307{
2308 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2309}
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
2316#endif
2317
2318eio_req *eio_grp (eio_cb cb, void *data)
2319{
2320 const int pri = EIO_PRI_MAX;
2321
2322 REQ (EIO_GROUP); SEND;
2323}
2324
2325#undef REQ
2326#undef PATH
2327#undef SEND
2328
2329/*****************************************************************************/
2330/* grp functions */
2331
1046void 2332void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 {
1130 dREQ;
1131
1132 req->type = ix;
1133 req->sv1 = newSVsv (fh);
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141
1142 if (!SvREADONLY (data))
1143 {
1144 SvREADONLY_on (data);
1145 req->flags |= FLAG_SV2_RO_OFF;
1146 }
1147
1148 EIO_SEND;
1149 }
1150}
1151
1152void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159
1160 data = newSV (NAME_MAX);
1161 SvPOK_on (data);
1162
1163 req->type = EIO_READLINK;
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170}
1171
1172void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2333eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2334{
1421 grp->int2 = limit; 2335 grp->int2 = limit;
1422 grp->feed = feed; 2336 grp->feed = feed;
1423 2337
1424 grp_try_feed (grp); 2338 grp_try_feed (grp);
1425} 2339}
1426 2340
2341void
1427void eio_grp_limit (eio_req *grp, int limit) 2342eio_grp_limit (eio_req *grp, int limit)
1428{ 2343{
1429 grp->int2 = limit; 2344 grp->int2 = limit;
1430 2345
1431 grp_try_feed (grp); 2346 grp_try_feed (grp);
1432} 2347}
1433 2348
2349void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2350eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2351{
1436 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;
1437 2355
1438 ++grp->size; 2356 ++grp->size;
1439 req->grp = grp; 2357 req->grp = grp;
1440 2358
1441 req->grp_prev = 0; 2359 req->grp_prev = 0;
1445 grp->grp_first->grp_prev = req; 2363 grp->grp_first->grp_prev = req;
1446 2364
1447 grp->grp_first = req; 2365 grp->grp_first = req;
1448} 2366}
1449 2367
2368/*****************************************************************************/
2369/* misc garbage */
1450 2370
2371eio_ssize_t
2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2373{
2374 etp_worker wrk;
2375 eio_ssize_t ret;
2376
2377 wrk.dbuf = 0;
2378
2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2380
2381 if (wrk.dbuf)
2382 free (wrk.dbuf);
2383
2384 return ret;
2385}
2386

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines