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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.89 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
165 #endif
166
43# endif 167#endif
44
45#endif
46
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50 168
51#if HAVE_SENDFILE 169#if HAVE_SENDFILE
52# if __linux 170# if __linux
53# include <sys/sendfile.h> 171# include <sys/sendfile.h>
54# elif __freebsd 172# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 173# include <sys/socket.h>
56# include <sys/uio.h> 174# include <sys/uio.h>
57# elif __hpux 175# elif __hpux
58# include <sys/socket.h> 176# include <sys/socket.h>
59# elif __solaris /* not yet */ 177# elif __solaris
60# include <sys/sendfile.h> 178# include <sys/sendfile.h>
61# else 179# else
62# error sendfile support requested but not available 180# error sendfile support requested but not available
63# endif 181# endif
64#endif 182#endif
65 183
66/* number of seconds after which an idle threads exit */ 184#ifndef D_TYPE
67#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
68 193
69/* used for struct dirent, AIX doesn't provide it */ 194/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 195#ifndef NAME_MAX
71# define NAME_MAX 4096 196# define NAME_MAX 4096
72#endif 197#endif
73 198
199/* used for readlink etc. */
200#ifndef PATH_MAX
201# define PATH_MAX 4096
202#endif
203
74/* buffer size for various temporary buffers */ 204/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 205#define EIO_BUFSIZE 65536
76 206
77#define dBUF \ 207#define dBUF \
78 char *eio_buf; \ 208 char *eio_buf; \
79 X_LOCK (wrklock); \ 209 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 210 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 211 ETP_WORKER_UNLOCK (self); \
212 errno = ENOMEM; \
82 if (!eio_buf) \ 213 if (!eio_buf) \
83 return -1; 214 return -1;
84 215
85#define EIO_TICKS ((1000000 + 1023) >> 10) 216#define EIO_TICKS ((1000000 + 1023) >> 10)
86 217
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
87static void (*want_poll_cb) (void); 261static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 262static void (*done_poll_cb) (void);
89 263
90static unsigned int max_poll_time = 0; 264static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 265static unsigned int max_poll_reqs; /* reslock */
92 266
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 267static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 268static volatile unsigned int nready; /* reqlock */
95{ 269static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 272
100static 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)
101 304
102/* worker threads management */ 305/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 306
105typedef struct worker { 307static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 308etp_worker_clear (etp_worker *wrk)
120{ 309{
121 if (wrk->dirp) 310 ETP_WORKER_CLEAR (wrk);
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126
127 if (wrk->dbuf)
128 {
129 free (wrk->dbuf);
130 wrk->dbuf = 0;
131 }
132} 311}
133 312
134static void worker_free (worker *wrk) 313static void ecb_cold
314etp_worker_free (etp_worker *wrk)
135{ 315{
136 wrk->next->prev = wrk->prev; 316 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 317 wrk->prev->next = wrk->next;
138 318
139 free (wrk); 319 free (wrk);
140} 320}
141 321
142static volatile unsigned int nreqs, nready, npending; 322static unsigned int
143static volatile unsigned int max_idle = 4; 323etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 324{
325 int retval;
326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
327 retval = nreqs;
328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 329 return retval;
152} 330}
153 331
154unsigned int eio_nready (void) 332static unsigned int
333etp_nready (void)
155{ 334{
156 unsigned int retval; 335 unsigned int retval;
157 336
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 338 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 340
162 return retval; 341 return retval;
163} 342}
164 343
165unsigned int eio_npending (void) 344static unsigned int
345etp_npending (void)
166{ 346{
167 unsigned int retval; 347 unsigned int retval;
168 348
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 350 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 352
173 return retval; 353 return retval;
174} 354}
175 355
176unsigned int eio_nthreads (void) 356static unsigned int
357etp_nthreads (void)
177{ 358{
178 unsigned int retval; 359 unsigned int retval;
179 360
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 362 retval = started;
188 * a somewhat faster data structure might be nice, but 369 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 370 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 371 * per shift, the most expensive operation.
191 */ 372 */
192typedef struct { 373typedef struct {
193 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 */
194 int size; 375 int size;
195} reqq; 376} etp_reqq;
196 377
197static reqq req_queue; 378static etp_reqq req_queue;
198static reqq res_queue; 379static etp_reqq res_queue;
199 380
200static int reqq_push (reqq *q, eio_req *req) 381static int ecb_noinline
382reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 383{
202 int pri = req->pri; 384 int pri = req->pri;
203 req->next = 0; 385 req->next = 0;
204 386
205 if (q->qe[pri]) 387 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 393 q->qe[pri] = q->qs[pri] = req;
212 394
213 return q->size++; 395 return q->size++;
214} 396}
215 397
216static eio_req *reqq_shift (reqq *q) 398static ETP_REQ * ecb_noinline
399reqq_shift (etp_reqq *q)
217{ 400{
218 int pri; 401 int pri;
219 402
220 if (!q->size) 403 if (!q->size)
221 return 0; 404 return 0;
222 405
223 --q->size; 406 --q->size;
224 407
225 for (pri = EIO_NUM_PRI; pri--; ) 408 for (pri = ETP_NUM_PRI; pri--; )
226 { 409 {
227 eio_req *req = q->qs[pri]; 410 eio_req *req = q->qs[pri];
228 411
229 if (req) 412 if (req)
230 { 413 {
236 } 419 }
237 420
238 abort (); 421 abort ();
239} 422}
240 423
241static void grp_try_feed (eio_req *grp) 424static void ecb_cold
425etp_thread_init (void)
242{ 426{
243 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)
244 { 458 {
245 int old_len = grp->size; 459 etp_worker *wrk = wrk_first.next;
246 460
247 EIO_FEED (grp); 461 if (wrk->req)
462 ETP_DESTROY (wrk->req);
248 463
249 /* stop if no progress has been made */ 464 etp_worker_clear (wrk);
250 if (old_len == grp->size) 465 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 466 }
256}
257 467
258static int eio_finish (eio_req *req); 468 started = 0;
469 idle = 0;
470 nreqs = 0;
471 nready = 0;
472 npending = 0;
259 473
260static int grp_dec (eio_req *grp) 474 etp_thread_init ();
261{ 475}
262 --grp->size;
263 476
264 /* call feeder, if applicable */ 477static void ecb_cold
265 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}
266 483
267 /* finish, if done */ 484static int ecb_cold
268 if (!grp->size && grp->int1) 485etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 486{
270 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
271 return 0; 494 return 0;
272} 495}
273 496
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 497X_THREAD_PROC (etp_proc);
325 498
326static void start_thread (void) 499static void ecb_cold
500etp_start_thread (void)
327{ 501{
328 worker *wrk = calloc (1, sizeof (worker)); 502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 503
330 /*TODO*/ 504 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 505 assert (("unable to allocate worker thread data", wrk));
332 506
333 X_LOCK (wrklock); 507 X_LOCK (wrklock);
334 508
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 509 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 510 {
337 wrk->prev = &wrk_first; 511 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 512 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 513 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 514 wrk_first.next = wrk;
344 free (wrk); 518 free (wrk);
345 519
346 X_UNLOCK (wrklock); 520 X_UNLOCK (wrklock);
347} 521}
348 522
523static void
349static void maybe_start_thread (void) 524etp_maybe_start_thread (void)
350{ 525{
351 if (eio_nthreads () >= wanted) 526 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 527 return;
353 528
354 /* todo: maybe use idle here, but might be less exact */ 529 /* todo: maybe use idle here, but might be less exact */
355 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 ()))
356 return; 531 return;
357 532
358 start_thread (); 533 etp_start_thread ();
359} 534}
360 535
361void eio_submit (eio_req *req) 536static void ecb_cold
362{ 537etp_end_thread (void)
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void)
375{ 538{
376 eio_req *req = calloc (1, sizeof (eio_req)); 539 eio_req *req = calloc (1, sizeof (eio_req));
377 540
378 req->type = EIO_QUIT; 541 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 543
381 X_LOCK (reqlock); 544 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 545 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 546 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 547 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 549 X_LOCK (wrklock);
387 --started; 550 --started;
388 X_UNLOCK (wrklock); 551 X_UNLOCK (wrklock);
389} 552}
390 553
391void eio_set_max_poll_time (double nseconds) 554static int
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 555etp_poll (void)
428{ 556{
429 int maxreqs = max_poll_reqs; 557 unsigned int maxreqs;
558 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 559 struct timeval tv_start, tv_now;
431 eio_req *req;
432 560
561 X_LOCK (reslock);
562 maxreqs = max_poll_reqs;
563 maxtime = max_poll_time;
564 X_UNLOCK (reslock);
565
433 if (max_poll_time) 566 if (maxtime)
434 gettimeofday (&tv_start, 0); 567 gettimeofday (&tv_start, 0);
435 568
436 for (;;) 569 for (;;)
437 { 570 {
571 ETP_REQ *req;
572
438 maybe_start_thread (); 573 etp_maybe_start_thread ();
439 574
440 X_LOCK (reslock); 575 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 576 req = reqq_shift (&res_queue);
442 577
443 if (req) 578 if (req)
451 X_UNLOCK (reslock); 586 X_UNLOCK (reslock);
452 587
453 if (!req) 588 if (!req)
454 return 0; 589 return 0;
455 590
591 X_LOCK (reqlock);
456 --nreqs; 592 --nreqs;
593 X_UNLOCK (reqlock);
457 594
458 if (req->type == EIO_GROUP && req->size) 595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 596 {
460 req->int1 = 1; /* mark request as delayed */ 597 req->int1 = 1; /* mark request as delayed */
461 continue; 598 continue;
462 } 599 }
463 else 600 else
464 { 601 {
465 int res = eio_finish (req); 602 int res = ETP_FINISH (req);
466 if (res) 603 if (ecb_expect_false (res))
467 return res; 604 return res;
468 } 605 }
469 606
470 if (maxreqs && !--maxreqs) 607 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 608 break;
472 609
473 if (max_poll_time) 610 if (maxtime)
474 { 611 {
475 gettimeofday (&tv_now, 0); 612 gettimeofday (&tv_now, 0);
476 613
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 614 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 615 break;
479 } 616 }
480 } 617 }
481 618
482 errno = EAGAIN; 619 errno = EAGAIN;
483 return -1; 620 return -1;
484} 621}
485 622
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
486/*****************************************************************************/ 873/*****************************************************************************/
487/* work around various missing functions */ 874/* work around various missing functions */
488 875
489#if !HAVE_PREADWRITE 876#if !HAVE_PREADWRITE
877# undef pread
878# undef pwrite
490# define pread aio_pread 879# define pread eio__pread
491# define pwrite aio_pwrite 880# define pwrite eio__pwrite
492 881
493/* 882static eio_ssize_t
494 * make our pread/pwrite safe against themselves, but not against 883eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 884{
502 ssize_t res; 885 eio_ssize_t res;
503 off_t ooffset; 886 off_t ooffset;
504 887
505 X_LOCK (preadwritelock); 888 X_LOCK (preadwritelock);
506 ooffset = lseek (fd, 0, SEEK_CUR); 889 ooffset = lseek (fd, 0, SEEK_CUR);
507 lseek (fd, offset, SEEK_SET); 890 lseek (fd, offset, SEEK_SET);
510 X_UNLOCK (preadwritelock); 893 X_UNLOCK (preadwritelock);
511 894
512 return res; 895 return res;
513} 896}
514 897
515static 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)
516{ 900{
517 ssize_t res; 901 eio_ssize_t res;
518 off_t ooffset; 902 off_t ooffset;
519 903
520 X_LOCK (preadwritelock); 904 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 905 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 906 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 907 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 908 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 909 X_UNLOCK (preadwritelock);
526 910
527 return res; 911 return res;
528} 912}
529#endif 913#endif
530 914
531#ifndef HAVE_FUTIMES 915#ifndef HAVE_UTIMES
532 916
917# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 918# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 919
920static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 921eio__utimes (const char *filename, const struct timeval times[2])
537{ 922{
538 if (times) 923 if (times)
539 { 924 {
540 struct utimbuf buf; 925 struct utimbuf buf;
541 926
546 } 931 }
547 else 932 else
548 return utime (filename, 0); 933 return utime (filename, 0);
549} 934}
550 935
936#endif
937
938#ifndef HAVE_FUTIMES
939
940# undef futimes
941# define futimes(fd,times) eio__futimes (fd, times)
942
943static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 944eio__futimes (int fd, const struct timeval tv[2])
552{ 945{
553 errno = ENOSYS; 946 errno = ENOSYS;
554 return -1; 947 return -1;
555} 948}
556 949
557#endif 950#endif
558 951
559#if !HAVE_FDATASYNC 952#if !HAVE_FDATASYNC
953# undef fdatasync
560# define fdatasync fsync 954# define fdatasync(fd) fsync (fd)
561#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}
562 995
563#if !HAVE_READAHEAD 996#if !HAVE_READAHEAD
997# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 998# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 999
1000static eio_ssize_t
566static 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)
567{ 1002{
568 size_t todo = count; 1003 size_t todo = count;
569 dBUF; 1004 dBUF;
570 1005
571 while (todo > 0) 1006 while (todo > 0)
581 return count; 1016 return count;
582} 1017}
583 1018
584#endif 1019#endif
585 1020
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif
615
616/* sendfile always needs emulation */ 1021/* sendfile always needs emulation */
1022static eio_ssize_t
617static 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)
618{ 1024{
1025 eio_ssize_t written = 0;
619 ssize_t res; 1026 eio_ssize_t res;
620 1027
621 if (!count) 1028 if (!count)
622 return 0; 1029 return 0;
623 1030
1031 for (;;)
1032 {
1033#ifdef __APPLE__
1034# undef HAVE_SENDFILE /* broken, as everything on os x */
1035#endif
624#if HAVE_SENDFILE 1036#if HAVE_SENDFILE
625# if __linux 1037# if __linux
1038 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1039 res = sendfile (ofd, ifd, &soffset, count);
627 1040
628# elif __freebsd 1041# elif __FreeBSD__
629 /* 1042 /*
630 * 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
631 * wasted on making it similar to other I/O functions. 1044 * wasted on making it similar to other I/O functions.
632 */ 1045 */
633 {
634 off_t sbytes; 1046 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1047 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1048
637 if (res < 0 && sbytes) 1049 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* 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)
639 res = sbytes; 1054 res = sbytes;
640 } 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;
641 1068
642# elif __hpux 1069# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1070 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1071
645# elif __solaris 1072# elif __solaris
646 {
647 struct sendfilevec vec; 1073 struct sendfilevec vec;
648 size_t sbytes; 1074 size_t sbytes;
649 1075
650 vec.sfv_fd = ifd; 1076 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1077 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1078 vec.sfv_off = offset;
653 vec.sfv_len = count; 1079 vec.sfv_len = count;
654 1080
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1081 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1082
657 if (res < 0 && sbytes) 1083 if (res < 0 && sbytes)
658 res = sbytes; 1084 res = sbytes;
659 }
660 1085
661# 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
662#else 1097#else
663 res = -1; 1098 res = -1;
664 errno = ENOSYS; 1099 errno = ENOSYS;
665#endif 1100#endif
666 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
667 if (res < 0 1125 if (res < 0
668 && (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 */
669#if __solaris 1132#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1134#endif
672 ) 1135 )
673 ) 1136 )
677 1140
678 res = 0; 1141 res = 0;
679 1142
680 while (count) 1143 while (count)
681 { 1144 {
682 ssize_t cnt; 1145 eio_ssize_t cnt;
683 1146
684 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);
685 1148
686 if (cnt <= 0) 1149 if (cnt <= 0)
687 { 1150 {
704 } 1167 }
705 1168
706 return res; 1169 return res;
707} 1170}
708 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
709/* read a full directory */ 1616/* read a full directory */
1617static void
710static void scandir_ (eio_req *req, worker *self) 1618eio__scandir (eio_req *req, etp_worker *self)
711{ 1619{
712 DIR *dirp; 1620 DIR *dirp;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1622 char *name, *names;
720 int memlen = 4096; 1623 int namesalloc = 4096;
721 int memofs = 0; 1624 int namesoffs = 0;
1625 int flags = req->int1;
1626 eio_dirent *dents = 0;
1627 int dentalloc = 128;
722 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);
723 1635
724 X_LOCK (wrklock); 1636 X_LOCK (wrklock);
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1638 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1639
727 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;
728 req->ptr2 = names = malloc (memlen); 1645 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1646 X_UNLOCK (wrklock);
730 1647
731 if (dirp && u && names) 1648 if (dirp && names && (!flags || dents))
732 for (;;) 1649 for (;;)
733 { 1650 {
734 errno = 0; 1651 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1652 entp = readdir (dirp);
736 1653
737 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
738 break; 1692 break;
1693 }
739 1694
1695 /* now add the entry to our list(s) */
740 name = entp->d_name; 1696 name = entp->d_name;
741 1697
1698 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1700 {
744 int len = strlen (name) + 1; 1701 int len = D_NAMLEN (entp) + 1;
745 1702
746 res++; 1703 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1704 {
750 memlen *= 2; 1705 namesalloc *= 2;
751 X_LOCK (wrklock); 1706 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1707 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1708 X_UNLOCK (wrklock);
754 1709
755 if (!names) 1710 if (!names)
756 break; 1711 break;
757 } 1712 }
758 1713
759 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
760 memofs += len; 1806 namesoffs += len;
1807 ++dentoffs;
1808 }
1809
1810 if (EIO_CANCELLED (req))
1811 {
1812 errno = ECANCELED;
1813 break;
761 } 1814 }
762 } 1815 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1816}
769 1817
770/*****************************************************************************/ 1818/*****************************************************************************/
771 1819
772#define ALLOC(len) \ 1820#define ALLOC(len) \
782 req->result = -1; \ 1830 req->result = -1; \
783 break; \ 1831 break; \
784 } \ 1832 } \
785 } 1833 }
786 1834
787X_THREAD_PROC (eio_proc) 1835X_THREAD_PROC (etp_proc)
788{ 1836{
789 eio_req *req; 1837 ETP_REQ *req;
790 struct timespec ts; 1838 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
792 1841
793 /* try to distribute timeouts somewhat randomly */ 1842 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1844
796 for (;;) 1845 for (;;)
797 { 1846 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1847 ts.tv_sec = 0;
799 1848
800 X_LOCK (reqlock); 1849 X_LOCK (reqlock);
801 1850
802 for (;;) 1851 for (;;)
803 { 1852 {
804 self->req = req = reqq_shift (&req_queue); 1853 self->req = req = reqq_shift (&req_queue);
805 1854
806 if (req) 1855 if (req)
807 break; 1856 break;
808 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
809 ++idle; 1867 ++idle;
810 1868
811 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
812 { 1873 {
813 if (idle > max_idle) 1874 /* initialise timeout once */
814 { 1875 if (!ts.tv_sec)
815 --idle; 1876 ts.tv_sec = time (0) + idle_timeout;
816 X_UNLOCK (reqlock);
817 X_LOCK (wrklock);
818 --started;
819 X_UNLOCK (wrklock);
820 goto quit;
821 }
822 1877
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 1880 }
827 1881
828 --idle; 1882 --idle;
829 } 1883 }
830 1884
831 --nready; 1885 --nready;
832 1886
833 X_UNLOCK (reqlock); 1887 X_UNLOCK (reqlock);
834 1888
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 1889 if (req->type < 0)
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit; 1890 goto quit;
930 1891
931 default: 1892 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1893
938 X_LOCK (reslock); 1894 X_LOCK (reslock);
939 1895
940 ++npending; 1896 ++npending;
941 1897
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1898 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1899 want_poll_cb ();
944 1900
945 self->req = 0; 1901 self->req = 0;
946 worker_clear (self); 1902 etp_worker_clear (self);
947 1903
948 X_UNLOCK (reslock); 1904 X_UNLOCK (reslock);
949 } 1905 }
950 1906
951quit: 1907quit:
952 X_LOCK (wrklock); 1908 X_LOCK (wrklock);
953 worker_free (self); 1909 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1910 X_UNLOCK (wrklock);
955 1911
956 return 0; 1912 return 0;
957} 1913}
958 1914
959/*****************************************************************************/ 1915/*****************************************************************************/
960 1916
961static void eio_atfork_prepare (void) 1917int ecb_cold
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1918eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1919{
1019 want_poll_cb = want_poll; 1920 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 1921}
1035 1922
1923ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1924eio_api_destroy (eio_req *req)
1037{ 1925{
1038 free (req); 1926 free (req);
1039} 1927}
1040 1928
1041#define REQ(rtype) \ 1929#define REQ(rtype) \
1043 \ 1931 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1932 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1933 if (!req) \
1046 return 0; \ 1934 return 0; \
1047 \ 1935 \
1048 req->type = EIO_ ## rtype; \ 1936 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1937 req->pri = pri; \
1050 req->finish = cb; \ 1938 req->finish = cb; \
1939 req->data = data; \
1051 req->destroy = eio_api_destroy; 1940 req->destroy = eio_api_destroy;
1052 1941
1053#define SEND eio_submit (req); return req 1942#define SEND eio_submit (req); return req
1054 1943
1055#define PATH (void)0 1944#define PATH \
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 }
1056 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
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
1057eio_req *eio_fsync (int fd, eio_cb cb) 2094eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2095{
2096 REQ (EIO_SYNC); SEND;
2097}
2098
2099eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2100{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2101 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2102}
1061 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
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2135{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2137}
1066 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
1067eio_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)
1068{ 2145{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2146 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2147}
1071 2148
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2149eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2150{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2151 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2152}
1076 2153
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2154eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2155{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2156 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2157}
1081 2158
1082eio_req *eio_fstat (int fd, eio_cb cb) 2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2160{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2161 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2162}
1086 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
1087eio_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)
1088{ 2170{
1089 REQ (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;
1090} 2172}
1091 2173
1092eio_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)
1093{ 2175{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2176 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2177}
1096 2178
1097eio_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)
1098{ 2180{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2181 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2182}
1101 2183
1102eio_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)
1103{ 2185{
1104 REQ (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;
1105} 2187}
1106 2188
1107eio_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)
1108{ 2190{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2191 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2192}
1111 2193
1112eio_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)
1113{ 2195{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2196 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2197}
1116 2198
1117eio_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)
1118{ 2200{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2201 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2202}
1121 2203
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2204eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2205{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2206 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2207}
1141 2208
1142eio_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)
1143{ 2210{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2211 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2212}
1146 2213
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2214eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2215{
1149 REQ (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;
1150} 2217}
1151 2218
1152eio_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)
1153{ 2220{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2221 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2222}
1156 2223
1157eio_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)
1158{ 2225{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2226 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2227}
1161 2228
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2229static eio_req *
2230eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2231{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2232 REQ (type); PATH; SEND;
1170} 2233}
1171 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
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
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2265eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2266{
1174 REQ (READDIR); PATH; SEND; 2267 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2268}
1176 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
1177eio_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)
1178{ 2276{
1179 REQ (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;
1180} 2278}
1181 2279
1182eio_req *eio_busy (double delay, eio_cb cb) 2280static eio_req *
2281eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2282{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2283 REQ (type); PATH;
1185}
1186 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
1187eio_req *eio_nop (eio_cb cb) 2318eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2319{
2320 const int pri = EIO_PRI_MAX;
2321
1189 REQ (NOP); SEND; 2322 REQ (EIO_GROUP); SEND;
1190} 2323}
1191 2324
1192#undef REQ 2325#undef REQ
1193#undef PATH 2326#undef PATH
1194#undef SEND 2327#undef SEND
1195 2328
1196#if 0 2329/*****************************************************************************/
2330/* grp functions */
1197 2331
1198void 2332void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2333eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2334{
1573 grp->int2 = limit; 2335 grp->int2 = limit;
1574 grp->feed = feed; 2336 grp->feed = feed;
1575 2337
1576 grp_try_feed (grp); 2338 grp_try_feed (grp);
1577} 2339}
1578 2340
2341void
1579void eio_grp_limit (eio_req *grp, int limit) 2342eio_grp_limit (eio_req *grp, int limit)
1580{ 2343{
1581 grp->int2 = limit; 2344 grp->int2 = limit;
1582 2345
1583 grp_try_feed (grp); 2346 grp_try_feed (grp);
1584} 2347}
1585 2348
2349void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2350eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2351{
1588 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;
1589 2355
1590 ++grp->size; 2356 ++grp->size;
1591 req->grp = grp; 2357 req->grp = grp;
1592 2358
1593 req->grp_prev = 0; 2359 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2363 grp->grp_first->grp_prev = req;
1598 2364
1599 grp->grp_first = req; 2365 grp->grp_first = req;
1600} 2366}
1601 2367
2368/*****************************************************************************/
2369/* misc garbage */
1602 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