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 1.86 by root, Thu Jul 14 18:30:10 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63#include <sys/statvfs.h>
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
75
15#ifndef EIO_FINISH 76#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 78#endif
18 79
19#ifndef EIO_DESTROY 80#ifndef EIO_DESTROY
22 83
23#ifndef EIO_FEED 84#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 85# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 86#endif
26 87
88#ifndef EIO_FD_TO_WIN32_HANDLE
89# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
90#endif
91#ifndef EIO_WIN32_HANDLE_TO_FD
92# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
93#endif
94
95#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
96
97#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98
27#ifdef _WIN32 99#ifdef _WIN32
28 100
29 /*doh*/ 101 #define PAGESIZE 4096 /* GetSystemInfo? */
102
103 #define stat(path,buf) _stati64 (path,buf)
104 #define lstat(path,buf) stat (path,buf)
105 #define fstat(fd,buf) _fstati64 (path,buf)
106 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
107 #define mkdir(path,mode) _mkdir (path)
108 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
109
110 #define chown(path,uid,gid) EIO_ENOSYS ()
111 #define fchown(fd,uid,gid) EIO_ENOSYS ()
112 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
113 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
114 #define mknod(path,mode,dev) EIO_ENOSYS ()
115 #define sync() EIO_ENOSYS ()
116
117 /* we could even stat and see if it exists */
118 static int
119 symlink (const char *old, const char *neu)
120 {
121 if (CreateSymbolicLink (neu, old, 1))
122 return 0;
123
124 if (CreateSymbolicLink (neu, old, 0))
125 return 0;
126
127 return EIO_ERRNO (ENOENT, -1);
128 }
30 129
31#else 130#else
32 131
33# include "config.h"
34# include <sys/time.h> 132 #include <sys/time.h>
35# include <sys/select.h> 133 #include <sys/select.h>
134 #include <sys/statvfs.h>
36# include <unistd.h> 135 #include <unistd.h>
37# include <utime.h> 136 #include <utime.h>
38# include <signal.h> 137 #include <signal.h>
39# include <dirent.h> 138 #include <dirent.h>
40 139
140 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
141 #include <sys/mman.h>
142 #endif
143
144 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
145 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
146 #define _DIRENT_HAVE_D_TYPE /* sigh */
147 #define D_INO(de) (de)->d_fileno
148 #define D_NAMLEN(de) (de)->d_namlen
149 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
150 #define D_INO(de) (de)->d_ino
151 #endif
152
153 #ifdef _D_EXACT_NAMLEN
154 #undef D_NAMLEN
155 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
156 #endif
157
158 #ifdef _DIRENT_HAVE_D_TYPE
159 #define D_TYPE(de) (de)->d_type
160 #endif
161
41# ifndef EIO_STRUCT_DIRENT 162 #ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 163 #define EIO_STRUCT_DIRENT struct dirent
164 #endif
165
43# endif 166#endif
44
45#endif
46
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50 167
51#if HAVE_SENDFILE 168#if HAVE_SENDFILE
52# if __linux 169# if __linux
53# include <sys/sendfile.h> 170# include <sys/sendfile.h>
54# elif __freebsd 171# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 172# include <sys/socket.h>
56# include <sys/uio.h> 173# include <sys/uio.h>
57# elif __hpux 174# elif __hpux
58# include <sys/socket.h> 175# include <sys/socket.h>
59# elif __solaris /* not yet */ 176# elif __solaris
60# include <sys/sendfile.h> 177# include <sys/sendfile.h>
61# else 178# else
62# error sendfile support requested but not available 179# error sendfile support requested but not available
63# endif 180# endif
64#endif 181#endif
65 182
66/* number of seconds after which an idle threads exit */ 183#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 184# define D_TYPE(de) 0
185#endif
186#ifndef D_INO
187# define D_INO(de) 0
188#endif
189#ifndef D_NAMLEN
190# define D_NAMLEN(de) strlen ((de)->d_name)
191#endif
68 192
69/* used for struct dirent, AIX doesn't provide it */ 193/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 194#ifndef NAME_MAX
71# define NAME_MAX 4096 195# define NAME_MAX 4096
72#endif 196#endif
73 197
198/* used for readlink etc. */
199#ifndef PATH_MAX
200# define PATH_MAX 4096
201#endif
202
74/* buffer size for various temporary buffers */ 203/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 204#define EIO_BUFSIZE 65536
76 205
77#define dBUF \ 206#define dBUF \
78 char *eio_buf; \ 207 char *eio_buf; \
79 X_LOCK (wrklock); \ 208 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 209 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 210 ETP_WORKER_UNLOCK (self); \
211 errno = ENOMEM; \
82 if (!eio_buf) \ 212 if (!eio_buf) \
83 return -1; 213 return -1;
84 214
85#define EIO_TICKS ((1000000 + 1023) >> 10) 215#define EIO_TICKS ((1000000 + 1023) >> 10)
86 216
217#define ETP_PRI_MIN EIO_PRI_MIN
218#define ETP_PRI_MAX EIO_PRI_MAX
219
220struct etp_worker;
221
222#define ETP_REQ eio_req
223#define ETP_DESTROY(req) eio_destroy (req)
224static int eio_finish (eio_req *req);
225#define ETP_FINISH(req) eio_finish (req)
226static void eio_execute (struct etp_worker *self, eio_req *req);
227#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
228
229#define ETP_WORKER_CLEAR(req) \
230 if (wrk->dbuf) \
231 { \
232 free (wrk->dbuf); \
233 wrk->dbuf = 0; \
234 } \
235 \
236 if (wrk->dirp) \
237 { \
238 closedir (wrk->dirp); \
239 wrk->dirp = 0; \
240 }
241
242#define ETP_WORKER_COMMON \
243 void *dbuf; \
244 DIR *dirp;
245
246/*****************************************************************************/
247
248#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
249
250/* calculate time difference in ~1/EIO_TICKS of a second */
251ecb_inline int
252tvdiff (struct timeval *tv1, struct timeval *tv2)
253{
254 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
255 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
256}
257
258static unsigned int started, idle, wanted = 4;
259
87static void (*want_poll_cb) (void); 260static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 261static void (*done_poll_cb) (void);
89 262
90static unsigned int max_poll_time = 0; 263static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 264static unsigned int max_poll_reqs; /* reslock */
92 265
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 266static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 267static volatile unsigned int nready; /* reqlock */
95{ 268static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 269static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 270static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 271
100static unsigned int started, idle, wanted; 272static xmutex_t wrklock;
273static xmutex_t reslock;
274static xmutex_t reqlock;
275static xcond_t reqwait;
276
277#if !HAVE_PREADWRITE
278/*
279 * make our pread/pwrite emulation safe against themselves, but not against
280 * normal read/write by using a mutex. slows down execution a lot,
281 * but that's your problem, not mine.
282 */
283static xmutex_t preadwritelock = X_MUTEX_INIT;
284#endif
285
286typedef struct etp_worker
287{
288 /* locked by wrklock */
289 struct etp_worker *prev, *next;
290
291 xthread_t tid;
292
293 /* locked by reslock, reqlock or wrklock */
294 ETP_REQ *req; /* currently processed request */
295
296 ETP_WORKER_COMMON
297} etp_worker;
298
299static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
300
301#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
302#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 303
102/* worker threads management */ 304/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 305
105typedef struct worker { 306static 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) 307etp_worker_clear (etp_worker *wrk)
120{ 308{
121 if (wrk->dirp) 309 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} 310}
133 311
134static void worker_free (worker *wrk) 312static void ecb_cold
313etp_worker_free (etp_worker *wrk)
135{ 314{
136 wrk->next->prev = wrk->prev; 315 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 316 wrk->prev->next = wrk->next;
138 317
139 free (wrk); 318 free (wrk);
140} 319}
141 320
142static volatile unsigned int nreqs, nready, npending; 321static unsigned int
143static volatile unsigned int max_idle = 4; 322etp_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{ 323{
324 int retval;
325 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
326 retval = nreqs;
327 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 328 return retval;
152} 329}
153 330
154unsigned int eio_nready (void) 331static unsigned int
332etp_nready (void)
155{ 333{
156 unsigned int retval; 334 unsigned int retval;
157 335
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 336 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 337 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 338 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 339
162 return retval; 340 return retval;
163} 341}
164 342
165unsigned int eio_npending (void) 343static unsigned int
344etp_npending (void)
166{ 345{
167 unsigned int retval; 346 unsigned int retval;
168 347
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 348 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 349 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 350 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 351
173 return retval; 352 return retval;
174} 353}
175 354
176unsigned int eio_nthreads (void) 355static unsigned int
356etp_nthreads (void)
177{ 357{
178 unsigned int retval; 358 unsigned int retval;
179 359
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 361 retval = started;
188 * a somewhat faster data structure might be nice, but 368 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 369 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 370 * per shift, the most expensive operation.
191 */ 371 */
192typedef struct { 372typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 373 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 374 int size;
195} reqq; 375} etp_reqq;
196 376
197static reqq req_queue; 377static etp_reqq req_queue;
198static reqq res_queue; 378static etp_reqq res_queue;
199 379
200static int reqq_push (reqq *q, eio_req *req) 380static int ecb_noinline
381reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 382{
202 int pri = req->pri; 383 int pri = req->pri;
203 req->next = 0; 384 req->next = 0;
204 385
205 if (q->qe[pri]) 386 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 392 q->qe[pri] = q->qs[pri] = req;
212 393
213 return q->size++; 394 return q->size++;
214} 395}
215 396
216static eio_req *reqq_shift (reqq *q) 397static ETP_REQ * ecb_noinline
398reqq_shift (etp_reqq *q)
217{ 399{
218 int pri; 400 int pri;
219 401
220 if (!q->size) 402 if (!q->size)
221 return 0; 403 return 0;
222 404
223 --q->size; 405 --q->size;
224 406
225 for (pri = EIO_NUM_PRI; pri--; ) 407 for (pri = ETP_NUM_PRI; pri--; )
226 { 408 {
227 eio_req *req = q->qs[pri]; 409 eio_req *req = q->qs[pri];
228 410
229 if (req) 411 if (req)
230 { 412 {
236 } 418 }
237 419
238 abort (); 420 abort ();
239} 421}
240 422
241static void grp_try_feed (eio_req *grp) 423static void ecb_cold
424etp_thread_init (void)
242{ 425{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 426#if !HAVE_PREADWRITE
427 X_MUTEX_CREATE (preadwritelock);
428#endif
429 X_MUTEX_CREATE (wrklock);
430 X_MUTEX_CREATE (reslock);
431 X_MUTEX_CREATE (reqlock);
432 X_COND_CREATE (reqwait);
433}
434
435static void ecb_cold
436etp_atfork_prepare (void)
437{
438}
439
440static void ecb_cold
441etp_atfork_parent (void)
442{
443}
444
445static void ecb_cold
446etp_atfork_child (void)
447{
448 ETP_REQ *prv;
449
450 while ((prv = reqq_shift (&req_queue)))
451 ETP_DESTROY (prv);
452
453 while ((prv = reqq_shift (&res_queue)))
454 ETP_DESTROY (prv);
455
456 while (wrk_first.next != &wrk_first)
244 { 457 {
245 int old_len = grp->size; 458 etp_worker *wrk = wrk_first.next;
246 459
247 EIO_FEED (grp); 460 if (wrk->req)
461 ETP_DESTROY (wrk->req);
248 462
249 /* stop if no progress has been made */ 463 etp_worker_clear (wrk);
250 if (old_len == grp->size) 464 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 465 }
256}
257 466
258static int eio_finish (eio_req *req); 467 started = 0;
468 idle = 0;
469 nreqs = 0;
470 nready = 0;
471 npending = 0;
259 472
260static int grp_dec (eio_req *grp) 473 etp_thread_init ();
261{ 474}
262 --grp->size;
263 475
264 /* call feeder, if applicable */ 476static void ecb_cold
265 grp_try_feed (grp); 477etp_once_init (void)
478{
479 etp_thread_init ();
480 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
481}
266 482
267 /* finish, if done */ 483static int ecb_cold
268 if (!grp->size && grp->int1) 484etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 485{
270 else 486 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
487
488 pthread_once (&doinit, etp_once_init);
489
490 want_poll_cb = want_poll;
491 done_poll_cb = done_poll;
492
271 return 0; 493 return 0;
272} 494}
273 495
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); 496X_THREAD_PROC (etp_proc);
325 497
326static void start_thread (void) 498static void ecb_cold
499etp_start_thread (void)
327{ 500{
328 worker *wrk = calloc (1, sizeof (worker)); 501 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 502
330 /*TODO*/ 503 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 504 assert (("unable to allocate worker thread data", wrk));
332 505
333 X_LOCK (wrklock); 506 X_LOCK (wrklock);
334 507
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 508 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 509 {
337 wrk->prev = &wrk_first; 510 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 511 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 512 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 513 wrk_first.next = wrk;
344 free (wrk); 517 free (wrk);
345 518
346 X_UNLOCK (wrklock); 519 X_UNLOCK (wrklock);
347} 520}
348 521
522static void
349static void maybe_start_thread (void) 523etp_maybe_start_thread (void)
350{ 524{
351 if (eio_nthreads () >= wanted) 525 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 526 return;
353 527
354 /* todo: maybe use idle here, but might be less exact */ 528 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 529 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 530 return;
357 531
358 start_thread (); 532 etp_start_thread ();
359} 533}
360 534
361void eio_submit (eio_req *req) 535static void ecb_cold
362{ 536etp_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{ 537{
376 eio_req *req = calloc (1, sizeof (eio_req)); 538 eio_req *req = calloc (1, sizeof (eio_req));
377 539
378 req->type = EIO_QUIT; 540 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 541 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 542
381 X_LOCK (reqlock); 543 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 544 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 545 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 546 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 548 X_LOCK (wrklock);
387 --started; 549 --started;
388 X_UNLOCK (wrklock); 550 X_UNLOCK (wrklock);
389} 551}
390 552
391void eio_set_max_poll_time (double nseconds) 553static 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) 554etp_poll (void)
428{ 555{
429 int maxreqs = max_poll_reqs; 556 unsigned int maxreqs;
557 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 558 struct timeval tv_start, tv_now;
431 eio_req *req;
432 559
560 X_LOCK (reslock);
561 maxreqs = max_poll_reqs;
562 maxtime = max_poll_time;
563 X_UNLOCK (reslock);
564
433 if (max_poll_time) 565 if (maxtime)
434 gettimeofday (&tv_start, 0); 566 gettimeofday (&tv_start, 0);
435 567
436 for (;;) 568 for (;;)
437 { 569 {
570 ETP_REQ *req;
571
438 maybe_start_thread (); 572 etp_maybe_start_thread ();
439 573
440 X_LOCK (reslock); 574 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 575 req = reqq_shift (&res_queue);
442 576
443 if (req) 577 if (req)
451 X_UNLOCK (reslock); 585 X_UNLOCK (reslock);
452 586
453 if (!req) 587 if (!req)
454 return 0; 588 return 0;
455 589
590 X_LOCK (reqlock);
456 --nreqs; 591 --nreqs;
592 X_UNLOCK (reqlock);
457 593
458 if (req->type == EIO_GROUP && req->size) 594 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 595 {
460 req->int1 = 1; /* mark request as delayed */ 596 req->int1 = 1; /* mark request as delayed */
461 continue; 597 continue;
462 } 598 }
463 else 599 else
464 { 600 {
465 int res = eio_finish (req); 601 int res = ETP_FINISH (req);
466 if (res) 602 if (ecb_expect_false (res))
467 return res; 603 return res;
468 } 604 }
469 605
470 if (maxreqs && !--maxreqs) 606 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 607 break;
472 608
473 if (max_poll_time) 609 if (maxtime)
474 { 610 {
475 gettimeofday (&tv_now, 0); 611 gettimeofday (&tv_now, 0);
476 612
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 613 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 614 break;
479 } 615 }
480 } 616 }
481 617
482 errno = EAGAIN; 618 errno = EAGAIN;
483 return -1; 619 return -1;
484} 620}
485 621
622static void
623etp_cancel (ETP_REQ *req)
624{
625 req->cancelled = 1;
626
627 eio_grp_cancel (req);
628}
629
630static void
631etp_submit (ETP_REQ *req)
632{
633 req->pri -= ETP_PRI_MIN;
634
635 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
636 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
637
638 if (ecb_expect_false (req->type == EIO_GROUP))
639 {
640 /* I hope this is worth it :/ */
641 X_LOCK (reqlock);
642 ++nreqs;
643 X_UNLOCK (reqlock);
644
645 X_LOCK (reslock);
646
647 ++npending;
648
649 if (!reqq_push (&res_queue, req) && want_poll_cb)
650 want_poll_cb ();
651
652 X_UNLOCK (reslock);
653 }
654 else
655 {
656 X_LOCK (reqlock);
657 ++nreqs;
658 ++nready;
659 reqq_push (&req_queue, req);
660 X_COND_SIGNAL (reqwait);
661 X_UNLOCK (reqlock);
662
663 etp_maybe_start_thread ();
664 }
665}
666
667static void ecb_cold
668etp_set_max_poll_time (double nseconds)
669{
670 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
671 max_poll_time = nseconds * EIO_TICKS;
672 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
673}
674
675static void ecb_cold
676etp_set_max_poll_reqs (unsigned int maxreqs)
677{
678 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
679 max_poll_reqs = maxreqs;
680 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
681}
682
683static void ecb_cold
684etp_set_max_idle (unsigned int nthreads)
685{
686 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
687 max_idle = nthreads;
688 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
689}
690
691static void ecb_cold
692etp_set_idle_timeout (unsigned int seconds)
693{
694 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
695 idle_timeout = seconds;
696 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
697}
698
699static void ecb_cold
700etp_set_min_parallel (unsigned int nthreads)
701{
702 if (wanted < nthreads)
703 wanted = nthreads;
704}
705
706static void ecb_cold
707etp_set_max_parallel (unsigned int nthreads)
708{
709 if (wanted > nthreads)
710 wanted = nthreads;
711
712 while (started > wanted)
713 etp_end_thread ();
714}
715
716/*****************************************************************************/
717
718static void
719grp_try_feed (eio_req *grp)
720{
721 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
722 {
723 grp->flags &= ~EIO_FLAG_GROUPADD;
724
725 EIO_FEED (grp);
726
727 /* stop if no progress has been made */
728 if (!(grp->flags & EIO_FLAG_GROUPADD))
729 {
730 grp->feed = 0;
731 break;
732 }
733 }
734}
735
736static int
737grp_dec (eio_req *grp)
738{
739 --grp->size;
740
741 /* call feeder, if applicable */
742 grp_try_feed (grp);
743
744 /* finish, if done */
745 if (!grp->size && grp->int1)
746 return eio_finish (grp);
747 else
748 return 0;
749}
750
751static void
752eio_destroy (eio_req *req)
753{
754 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
755 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
756
757 EIO_DESTROY (req);
758}
759
760static int
761eio_finish (eio_req *req)
762{
763 int res = EIO_FINISH (req);
764
765 if (req->grp)
766 {
767 int res2;
768 eio_req *grp = req->grp;
769
770 /* unlink request */
771 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
772 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
773
774 if (grp->grp_first == req)
775 grp->grp_first = req->grp_next;
776
777 res2 = grp_dec (grp);
778
779 if (!res)
780 res = res2;
781 }
782
783 eio_destroy (req);
784
785 return res;
786}
787
788void
789eio_grp_cancel (eio_req *grp)
790{
791 for (grp = grp->grp_first; grp; grp = grp->grp_next)
792 eio_cancel (grp);
793}
794
795void
796eio_cancel (eio_req *req)
797{
798 etp_cancel (req);
799}
800
801void
802eio_submit (eio_req *req)
803{
804 etp_submit (req);
805}
806
807unsigned int
808eio_nreqs (void)
809{
810 return etp_nreqs ();
811}
812
813unsigned int
814eio_nready (void)
815{
816 return etp_nready ();
817}
818
819unsigned int
820eio_npending (void)
821{
822 return etp_npending ();
823}
824
825unsigned int ecb_cold
826eio_nthreads (void)
827{
828 return etp_nthreads ();
829}
830
831void ecb_cold
832eio_set_max_poll_time (double nseconds)
833{
834 etp_set_max_poll_time (nseconds);
835}
836
837void ecb_cold
838eio_set_max_poll_reqs (unsigned int maxreqs)
839{
840 etp_set_max_poll_reqs (maxreqs);
841}
842
843void ecb_cold
844eio_set_max_idle (unsigned int nthreads)
845{
846 etp_set_max_idle (nthreads);
847}
848
849void ecb_cold
850eio_set_idle_timeout (unsigned int seconds)
851{
852 etp_set_idle_timeout (seconds);
853}
854
855void ecb_cold
856eio_set_min_parallel (unsigned int nthreads)
857{
858 etp_set_min_parallel (nthreads);
859}
860
861void ecb_cold
862eio_set_max_parallel (unsigned int nthreads)
863{
864 etp_set_max_parallel (nthreads);
865}
866
867int eio_poll (void)
868{
869 return etp_poll ();
870}
871
486/*****************************************************************************/ 872/*****************************************************************************/
487/* work around various missing functions */ 873/* work around various missing functions */
488 874
489#if !HAVE_PREADWRITE 875#if !HAVE_PREADWRITE
876# undef pread
877# undef pwrite
490# define pread aio_pread 878# define pread eio__pread
491# define pwrite aio_pwrite 879# define pwrite eio__pwrite
492 880
493/* 881static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 882eio__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{ 883{
502 ssize_t res; 884 ssize_t res;
503 off_t ooffset; 885 off_t ooffset;
504 886
505 X_LOCK (preadwritelock); 887 X_LOCK (preadwritelock);
510 X_UNLOCK (preadwritelock); 892 X_UNLOCK (preadwritelock);
511 893
512 return res; 894 return res;
513} 895}
514 896
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 897static ssize_t
898eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 899{
517 ssize_t res; 900 ssize_t res;
518 off_t ooffset; 901 off_t ooffset;
519 902
520 X_LOCK (preadwritelock); 903 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 904 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 905 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 906 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 907 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 908 X_UNLOCK (preadwritelock);
526 909
527 return res; 910 return res;
528} 911}
529#endif 912#endif
530 913
531#ifndef HAVE_FUTIMES 914#ifndef HAVE_UTIMES
532 915
916# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 917# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 918
919static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 920eio__utimes (const char *filename, const struct timeval times[2])
537{ 921{
538 if (times) 922 if (times)
539 { 923 {
540 struct utimbuf buf; 924 struct utimbuf buf;
541 925
546 } 930 }
547 else 931 else
548 return utime (filename, 0); 932 return utime (filename, 0);
549} 933}
550 934
935#endif
936
937#ifndef HAVE_FUTIMES
938
939# undef futimes
940# define futimes(fd,times) eio__futimes (fd, times)
941
942static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 943eio__futimes (int fd, const struct timeval tv[2])
552{ 944{
553 errno = ENOSYS; 945 errno = ENOSYS;
554 return -1; 946 return -1;
555} 947}
556 948
557#endif 949#endif
558 950
559#if !HAVE_FDATASYNC 951#if !HAVE_FDATASYNC
952# undef fdatasync
560# define fdatasync fsync 953# define fdatasync(fd) fsync (fd)
561#endif 954#endif
955
956/* sync_file_range always needs emulation */
957static int
958eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
959{
960#if HAVE_SYNC_FILE_RANGE
961 int res;
962
963 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
964 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
965 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
966 {
967 flags = 0
968 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
969 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
970 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
971 }
972
973 res = sync_file_range (fd, offset, nbytes, flags);
974
975 if (!res || errno != ENOSYS)
976 return res;
977#endif
978
979 /* even though we could play tricks with the flags, it's better to always
980 * call fdatasync, as that matches the expectation of its users best */
981 return fdatasync (fd);
982}
983
984static int
985eio__fallocate (int fd, int mode, off_t offset, size_t len)
986{
987#if HAVE_FALLOCATE
988 return fallocate (fd, mode, offset, len);
989#else
990 errno = ENOSYS;
991 return -1;
992#endif
993}
562 994
563#if !HAVE_READAHEAD 995#if !HAVE_READAHEAD
996# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 997# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 998
999static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1000eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 1001{
568 size_t todo = count; 1002 size_t todo = count;
569 dBUF; 1003 dBUF;
570 1004
571 while (todo > 0) 1005 while (todo > 0)
581 return count; 1015 return count;
582} 1016}
583 1017
584#endif 1018#endif
585 1019
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 */ 1020/* sendfile always needs emulation */
1021static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1022eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 1023{
1024 ssize_t written = 0;
619 ssize_t res; 1025 ssize_t res;
620 1026
621 if (!count) 1027 if (!count)
622 return 0; 1028 return 0;
623 1029
1030 for (;;)
1031 {
1032#ifdef __APPLE__
1033# undef HAVE_SENDFILE /* broken, as everything on os x */
1034#endif
624#if HAVE_SENDFILE 1035#if HAVE_SENDFILE
625# if __linux 1036# if __linux
1037 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1038 res = sendfile (ofd, ifd, &soffset, count);
627 1039
628# elif __freebsd 1040# elif __FreeBSD__
629 /* 1041 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 1042 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1043 * wasted on making it similar to other I/O functions.
632 */ 1044 */
633 {
634 off_t sbytes; 1045 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1046 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1047
637 if (res < 0 && sbytes) 1048 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1049 /* freebsd' sendfile will return 0 on success */
1050 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1051 /* not on e.g. EIO or EPIPE - sounds broken */
1052 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1053 res = sbytes;
640 } 1054 #endif
1055
1056 /* according to source inspection, this is correct, and useful behaviour */
1057 if (sbytes)
1058 res = sbytes;
1059
1060# elif defined (__APPLE__)
1061 off_t sbytes = count;
1062 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1063
1064 /* according to the manpage, sbytes is always valid */
1065 if (sbytes)
1066 res = sbytes;
641 1067
642# elif __hpux 1068# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1069 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1070
645# elif __solaris 1071# elif __solaris
646 {
647 struct sendfilevec vec; 1072 struct sendfilevec vec;
648 size_t sbytes; 1073 size_t sbytes;
649 1074
650 vec.sfv_fd = ifd; 1075 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1076 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1077 vec.sfv_off = offset;
653 vec.sfv_len = count; 1078 vec.sfv_len = count;
654 1079
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1080 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1081
657 if (res < 0 && sbytes) 1082 if (res < 0 && sbytes)
658 res = sbytes; 1083 res = sbytes;
659 }
660 1084
661# endif 1085# endif
1086
1087#elif defined (_WIN32) && 0
1088 /* does not work, just for documentation of what would need to be done */
1089 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1090 /* libeio guarantees that the file offset does not change, and windows */
1091 /* has no way to get an independent handle to the same file description */
1092 HANDLE h = TO_SOCKET (ifd);
1093 SetFilePointer (h, offset, 0, FILE_BEGIN);
1094 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1095
662#else 1096#else
663 res = -1; 1097 res = -1;
664 errno = ENOSYS; 1098 errno = ENOSYS;
665#endif 1099#endif
666 1100
1101 /* we assume sendfile can copy at least 128mb in one go */
1102 if (res <= 128 * 1024 * 1024)
1103 {
1104 if (res > 0)
1105 written += res;
1106
1107 if (written)
1108 return written;
1109
1110 break;
1111 }
1112 else
1113 {
1114 /* if we requested more, then probably the kernel was lazy */
1115 written += res;
1116 offset += res;
1117 count -= res;
1118
1119 if (!count)
1120 return written;
1121 }
1122 }
1123
667 if (res < 0 1124 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1125 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1126 /* BSDs */
1127#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1128 || errno == ENOTSUP
1129#endif
1130 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1131#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1132 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1133#endif
672 ) 1134 )
673 ) 1135 )
704 } 1166 }
705 1167
706 return res; 1168 return res;
707} 1169}
708 1170
1171#ifdef PAGESIZE
1172# define eio_pagesize() PAGESIZE
1173#else
1174static intptr_t
1175eio_pagesize (void)
1176{
1177 static intptr_t page;
1178
1179 if (!page)
1180 page = sysconf (_SC_PAGESIZE);
1181
1182 return page;
1183}
1184#endif
1185
1186static void
1187eio_page_align (void **addr, size_t *length)
1188{
1189 intptr_t mask = eio_pagesize () - 1;
1190
1191 /* round down addr */
1192 intptr_t adj = mask & (intptr_t)*addr;
1193
1194 *addr = (void *)((intptr_t)*addr - adj);
1195 *length += adj;
1196
1197 /* round up length */
1198 *length = (*length + mask) & ~mask;
1199}
1200
1201#if !_POSIX_MEMLOCK
1202# define eio__mlockall(a) eio_nosyscall()
1203#else
1204
1205static int
1206eio__mlockall (int flags)
1207{
1208 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1209 extern int mallopt (int, int);
1210 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1211 #endif
1212
1213 if (EIO_MCL_CURRENT != MCL_CURRENT
1214 || EIO_MCL_FUTURE != MCL_FUTURE)
1215 {
1216 flags = 0
1217 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1218 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1219 }
1220
1221 return mlockall (flags);
1222}
1223#endif
1224
1225#if !_POSIX_MEMLOCK_RANGE
1226# define eio__mlock(a,b) EIO_ENOSYS ()
1227#else
1228
1229static int
1230eio__mlock (void *addr, size_t length)
1231{
1232 eio_page_align (&addr, &length);
1233
1234 return mlock (addr, length);
1235}
1236
1237#endif
1238
1239#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1240# define eio__msync(a,b,c) EIO_ENOSYS ()
1241#else
1242
1243static int
1244eio__msync (void *mem, size_t len, int flags)
1245{
1246 eio_page_align (&mem, &len);
1247
1248 if (EIO_MS_ASYNC != MS_SYNC
1249 || EIO_MS_INVALIDATE != MS_INVALIDATE
1250 || EIO_MS_SYNC != MS_SYNC)
1251 {
1252 flags = 0
1253 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1254 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1255 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1256 }
1257
1258 return msync (mem, len, flags);
1259}
1260
1261#endif
1262
1263static int
1264eio__mtouch (eio_req *req)
1265{
1266 void *mem = req->ptr2;
1267 size_t len = req->size;
1268 int flags = req->int1;
1269
1270 eio_page_align (&mem, &len);
1271
1272 {
1273 intptr_t addr = (intptr_t)mem;
1274 intptr_t end = addr + len;
1275 intptr_t page = eio_pagesize ();
1276
1277 if (addr < end)
1278 if (flags & EIO_MT_MODIFY) /* modify */
1279 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1280 else
1281 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1282 }
1283
1284 return 0;
1285}
1286
1287/*****************************************************************************/
1288/* requests implemented outside eio_execute, because they are so large */
1289
1290static void
1291eio__realpath (eio_req *req, etp_worker *self)
1292{
1293 char *rel = req->ptr1;
1294 char *res;
1295 char *tmp1, *tmp2;
1296#if SYMLOOP_MAX > 32
1297 int symlinks = SYMLOOP_MAX;
1298#else
1299 int symlinks = 32;
1300#endif
1301
1302 req->result = -1;
1303
1304 errno = EINVAL;
1305 if (!rel)
1306 return;
1307
1308 errno = ENOENT;
1309 if (!*rel)
1310 return;
1311
1312 if (!req->ptr2)
1313 {
1314 X_LOCK (wrklock);
1315 req->flags |= EIO_FLAG_PTR2_FREE;
1316 X_UNLOCK (wrklock);
1317 req->ptr2 = malloc (PATH_MAX * 3);
1318
1319 errno = ENOMEM;
1320 if (!req->ptr2)
1321 return;
1322 }
1323
1324 res = req->ptr2;
1325 tmp1 = res + PATH_MAX;
1326 tmp2 = tmp1 + PATH_MAX;
1327
1328#if 0 /* disabled, the musl way to do things is just too racy */
1329#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1330 /* on linux we may be able to ask the kernel */
1331 {
1332 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1333
1334 if (fd >= 0)
1335 {
1336 sprintf (tmp1, "/proc/self/fd/%d", fd);
1337 req->result = readlink (tmp1, res, PATH_MAX);
1338 close (fd);
1339
1340 /* here we should probably stat the open file and the disk file, to make sure they still match */
1341
1342 if (req->result > 0)
1343 goto done;
1344 }
1345 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1346 return;
1347 }
1348#endif
1349#endif
1350
1351 if (*rel != '/')
1352 {
1353 if (!getcwd (res, PATH_MAX))
1354 return;
1355
1356 if (res [1]) /* only use if not / */
1357 res += strlen (res);
1358 }
1359
1360 while (*rel)
1361 {
1362 ssize_t len, linklen;
1363 char *beg = rel;
1364
1365 while (*rel && *rel != '/')
1366 ++rel;
1367
1368 len = rel - beg;
1369
1370 if (!len) /* skip slashes */
1371 {
1372 ++rel;
1373 continue;
1374 }
1375
1376 if (beg [0] == '.')
1377 {
1378 if (len == 1)
1379 continue; /* . - nop */
1380
1381 if (beg [1] == '.' && len == 2)
1382 {
1383 /* .. - back up one component, if possible */
1384
1385 while (res != req->ptr2)
1386 if (*--res == '/')
1387 break;
1388
1389 continue;
1390 }
1391 }
1392
1393 errno = ENAMETOOLONG;
1394 if (res + 1 + len + 1 >= tmp1)
1395 return;
1396
1397 /* copy one component */
1398 *res = '/';
1399 memcpy (res + 1, beg, len);
1400
1401 /* zero-terminate, for readlink */
1402 res [len + 1] = 0;
1403
1404 /* now check if it's a symlink */
1405 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1406
1407 if (linklen < 0)
1408 {
1409 if (errno != EINVAL)
1410 return;
1411
1412 /* it's a normal directory. hopefully */
1413 res += len + 1;
1414 }
1415 else
1416 {
1417 /* yay, it was a symlink - build new path in tmp2 */
1418 int rellen = strlen (rel);
1419
1420 errno = ENAMETOOLONG;
1421 if (linklen + 1 + rellen >= PATH_MAX)
1422 return;
1423
1424 errno = ELOOP;
1425 if (!--symlinks)
1426 return;
1427
1428 if (*tmp1 == '/')
1429 res = req->ptr2; /* symlink resolves to an absolute path */
1430
1431 /* we need to be careful, as rel might point into tmp2 already */
1432 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1433 tmp2 [linklen] = '/';
1434 memcpy (tmp2, tmp1, linklen);
1435
1436 rel = tmp2;
1437 }
1438 }
1439
1440 /* special case for the lone root path */
1441 if (res == req->ptr2)
1442 *res++ = '/';
1443
1444 req->result = res - (char *)req->ptr2;
1445
1446done:
1447 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1448}
1449
1450static signed char
1451eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1452{
1453 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1454 : a->inode < b->inode ? -1
1455 : a->inode > b->inode ? 1
1456 : 0;
1457}
1458
1459#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1460
1461#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1462#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1463
1464static void
1465eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1466{
1467 unsigned char bits [9 + sizeof (ino_t) * 8];
1468 unsigned char *bit = bits;
1469
1470 assert (CHAR_BIT == 8);
1471 assert (sizeof (eio_dirent) * 8 < 256);
1472 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1473 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1474
1475 if (size <= EIO_SORT_FAST)
1476 return;
1477
1478 /* first prepare an array of bits to test in our radix sort */
1479 /* try to take endianness into account, as well as differences in ino_t sizes */
1480 /* inode_bits must contain all inodes ORed together */
1481 /* which is used to skip bits that are 0 everywhere, which is very common */
1482 {
1483 ino_t endianness;
1484 int i, j;
1485
1486 /* we store the byte offset of byte n into byte n of "endianness" */
1487 for (i = 0; i < sizeof (ino_t); ++i)
1488 ((unsigned char *)&endianness)[i] = i;
1489
1490 *bit++ = 0;
1491
1492 for (i = 0; i < sizeof (ino_t); ++i)
1493 {
1494 /* shifting off the byte offsets out of "endianness" */
1495 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1496 endianness >>= 8;
1497
1498 for (j = 0; j < 8; ++j)
1499 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1500 *bit++ = offs + j;
1501 }
1502
1503 for (j = 0; j < 8; ++j)
1504 if (score_bits & (1 << j))
1505 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1506 }
1507
1508 /* now actually do the sorting (a variant of MSD radix sort) */
1509 {
1510 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1511 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1512 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1513 int stk_idx = 0;
1514
1515 base_stk [stk_idx] = dents;
1516 end_stk [stk_idx] = dents + size;
1517 bit_stk [stk_idx] = bit - 1;
1518
1519 do
1520 {
1521 base = base_stk [stk_idx];
1522 end = end_stk [stk_idx];
1523 bit = bit_stk [stk_idx];
1524
1525 for (;;)
1526 {
1527 unsigned char O = *bit >> 3;
1528 unsigned char M = 1 << (*bit & 7);
1529
1530 eio_dirent *a = base;
1531 eio_dirent *b = end;
1532
1533 if (b - a < EIO_SORT_CUTOFF)
1534 break;
1535
1536 /* now bit-partition the array on the bit */
1537 /* this ugly asymmetric loop seems to perform much better than typical */
1538 /* partition algos found in the literature */
1539 do
1540 if (!(((unsigned char *)a)[O] & M))
1541 ++a;
1542 else if (!(((unsigned char *)--b)[O] & M))
1543 {
1544 eio_dirent tmp = *a; *a = *b; *b = tmp;
1545 ++a;
1546 }
1547 while (b > a);
1548
1549 /* next bit, or stop, if no bits left in this path */
1550 if (!*--bit)
1551 break;
1552
1553 base_stk [stk_idx] = a;
1554 end_stk [stk_idx] = end;
1555 bit_stk [stk_idx] = bit;
1556 ++stk_idx;
1557
1558 end = a;
1559 }
1560 }
1561 while (stk_idx--);
1562 }
1563}
1564
1565static void
1566eio_dent_insertion_sort (eio_dirent *dents, int size)
1567{
1568 /* first move the smallest element to the front, to act as a sentinel */
1569 {
1570 int i;
1571 eio_dirent *min = dents;
1572
1573 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1574 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1575 if (EIO_DENT_CMP (dents [i], <, *min))
1576 min = &dents [i];
1577
1578 /* swap elements 0 and j (minimum) */
1579 {
1580 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1581 }
1582 }
1583
1584 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1585 {
1586 eio_dirent *i, *j;
1587
1588 for (i = dents + 1; i < dents + size; ++i)
1589 {
1590 eio_dirent value = *i;
1591
1592 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1593 j [1] = j [0];
1594
1595 j [1] = value;
1596 }
1597 }
1598}
1599
1600static void
1601eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1602{
1603 if (size <= 1)
1604 return; /* our insertion sort relies on size > 0 */
1605
1606 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1607 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1608 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1609
1610 /* use an insertion sort at the end, or for small arrays, */
1611 /* as insertion sort is more efficient for small partitions */
1612 eio_dent_insertion_sort (dents, size);
1613}
1614
709/* read a full directory */ 1615/* read a full directory */
1616static void
710static void scandir_ (eio_req *req, worker *self) 1617eio__scandir (eio_req *req, etp_worker *self)
711{ 1618{
712 DIR *dirp; 1619 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; 1620 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1621 char *name, *names;
720 int memlen = 4096; 1622 int namesalloc = 4096;
721 int memofs = 0; 1623 int namesoffs = 0;
1624 int flags = req->int1;
1625 eio_dirent *dents = 0;
1626 int dentalloc = 128;
722 int res = 0; 1627 int dentoffs = 0;
1628 ino_t inode_bits = 0;
1629
1630 req->result = -1;
1631
1632 if (!(flags & EIO_READDIR_DENTS))
1633 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1634
724 X_LOCK (wrklock); 1635 X_LOCK (wrklock);
1636 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1637 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1638
727 req->flags |= EIO_FLAG_PTR2_FREE; 1639 if (req->flags & EIO_FLAG_PTR1_FREE)
1640 free (req->ptr1);
1641
1642 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1643 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1644 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1645 X_UNLOCK (wrklock);
730 1646
731 if (dirp && u && names) 1647 if (dirp && names && (!flags || dents))
732 for (;;) 1648 for (;;)
733 { 1649 {
734 errno = 0; 1650 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1651 entp = readdir (dirp);
736 1652
737 if (!entp) 1653 if (!entp)
1654 {
1655 if (errno)
1656 break;
1657
1658 /* sort etc. */
1659 req->int1 = flags;
1660 req->result = dentoffs;
1661
1662 if (flags & EIO_READDIR_STAT_ORDER)
1663 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1664 else if (flags & EIO_READDIR_DIRS_FIRST)
1665 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1666 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1667 else
1668 {
1669 /* in this case, all is known, and we just put dirs first and sort them */
1670 eio_dirent *oth = dents + dentoffs;
1671 eio_dirent *dir = dents;
1672
1673 /* now partition dirs to the front, and non-dirs to the back */
1674 /* by walking from both sides and swapping if necessary */
1675 while (oth > dir)
1676 {
1677 if (dir->type == EIO_DT_DIR)
1678 ++dir;
1679 else if ((--oth)->type == EIO_DT_DIR)
1680 {
1681 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1682
1683 ++dir;
1684 }
1685 }
1686
1687 /* now sort the dirs only (dirs all have the same score) */
1688 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1689 }
1690
738 break; 1691 break;
1692 }
739 1693
1694 /* now add the entry to our list(s) */
740 name = entp->d_name; 1695 name = entp->d_name;
741 1696
1697 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1698 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1699 {
744 int len = strlen (name) + 1; 1700 int len = D_NAMLEN (entp) + 1;
745 1701
746 res++; 1702 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1703 {
750 memlen *= 2; 1704 namesalloc *= 2;
751 X_LOCK (wrklock); 1705 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1706 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1707 X_UNLOCK (wrklock);
754 1708
755 if (!names) 1709 if (!names)
756 break; 1710 break;
757 } 1711 }
758 1712
759 memcpy (names + memofs, name, len); 1713 memcpy (names + namesoffs, name, len);
1714
1715 if (dents)
1716 {
1717 struct eio_dirent *ent;
1718
1719 if (ecb_expect_false (dentoffs == dentalloc))
1720 {
1721 dentalloc *= 2;
1722 X_LOCK (wrklock);
1723 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1724 X_UNLOCK (wrklock);
1725
1726 if (!dents)
1727 break;
1728 }
1729
1730 ent = dents + dentoffs;
1731
1732 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1733 ent->namelen = len - 1;
1734 ent->inode = D_INO (entp);
1735
1736 inode_bits |= ent->inode;
1737
1738 switch (D_TYPE (entp))
1739 {
1740 default:
1741 ent->type = EIO_DT_UNKNOWN;
1742 flags |= EIO_READDIR_FOUND_UNKNOWN;
1743 break;
1744
1745 #ifdef DT_FIFO
1746 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1747 #endif
1748 #ifdef DT_CHR
1749 case DT_CHR: ent->type = EIO_DT_CHR; break;
1750 #endif
1751 #ifdef DT_MPC
1752 case DT_MPC: ent->type = EIO_DT_MPC; break;
1753 #endif
1754 #ifdef DT_DIR
1755 case DT_DIR: ent->type = EIO_DT_DIR; break;
1756 #endif
1757 #ifdef DT_NAM
1758 case DT_NAM: ent->type = EIO_DT_NAM; break;
1759 #endif
1760 #ifdef DT_BLK
1761 case DT_BLK: ent->type = EIO_DT_BLK; break;
1762 #endif
1763 #ifdef DT_MPB
1764 case DT_MPB: ent->type = EIO_DT_MPB; break;
1765 #endif
1766 #ifdef DT_REG
1767 case DT_REG: ent->type = EIO_DT_REG; break;
1768 #endif
1769 #ifdef DT_NWK
1770 case DT_NWK: ent->type = EIO_DT_NWK; break;
1771 #endif
1772 #ifdef DT_CMP
1773 case DT_CMP: ent->type = EIO_DT_CMP; break;
1774 #endif
1775 #ifdef DT_LNK
1776 case DT_LNK: ent->type = EIO_DT_LNK; break;
1777 #endif
1778 #ifdef DT_SOCK
1779 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1780 #endif
1781 #ifdef DT_DOOR
1782 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1783 #endif
1784 #ifdef DT_WHT
1785 case DT_WHT: ent->type = EIO_DT_WHT; break;
1786 #endif
1787 }
1788
1789 ent->score = 7;
1790
1791 if (flags & EIO_READDIR_DIRS_FIRST)
1792 {
1793 if (ent->type == EIO_DT_UNKNOWN)
1794 {
1795 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1796 ent->score = 1;
1797 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1798 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1799 }
1800 else if (ent->type == EIO_DT_DIR)
1801 ent->score = 0;
1802 }
1803 }
1804
760 memofs += len; 1805 namesoffs += len;
1806 ++dentoffs;
1807 }
1808
1809 if (EIO_CANCELLED (req))
1810 {
1811 errno = ECANCELED;
1812 break;
761 } 1813 }
762 } 1814 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1815}
769 1816
770/*****************************************************************************/ 1817/*****************************************************************************/
771 1818
772#define ALLOC(len) \ 1819#define ALLOC(len) \
782 req->result = -1; \ 1829 req->result = -1; \
783 break; \ 1830 break; \
784 } \ 1831 } \
785 } 1832 }
786 1833
787X_THREAD_PROC (eio_proc) 1834X_THREAD_PROC (etp_proc)
788{ 1835{
789 eio_req *req; 1836 ETP_REQ *req;
790 struct timespec ts; 1837 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1838 etp_worker *self = (etp_worker *)thr_arg;
1839 int timeout;
792 1840
793 /* try to distribute timeouts somewhat randomly */ 1841 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1842 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1843
796 for (;;) 1844 for (;;)
797 { 1845 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1846 ts.tv_sec = 0;
799 1847
800 X_LOCK (reqlock); 1848 X_LOCK (reqlock);
801 1849
802 for (;;) 1850 for (;;)
803 { 1851 {
804 self->req = req = reqq_shift (&req_queue); 1852 self->req = req = reqq_shift (&req_queue);
805 1853
806 if (req) 1854 if (req)
807 break; 1855 break;
808 1856
1857 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1858 {
1859 X_UNLOCK (reqlock);
1860 X_LOCK (wrklock);
1861 --started;
1862 X_UNLOCK (wrklock);
1863 goto quit;
1864 }
1865
809 ++idle; 1866 ++idle;
810 1867
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1868 if (idle <= max_idle)
1869 /* we are allowed to idle, so do so without any timeout */
1870 X_COND_WAIT (reqwait, reqlock);
1871 else
812 { 1872 {
813 if (idle > max_idle) 1873 /* initialise timeout once */
814 { 1874 if (!ts.tv_sec)
815 --idle; 1875 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 1876
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1877 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1878 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 1879 }
827 1880
828 --idle; 1881 --idle;
829 } 1882 }
830 1883
831 --nready; 1884 --nready;
832 1885
833 X_UNLOCK (reqlock); 1886 X_UNLOCK (reqlock);
834 1887
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 1888 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; 1889 goto quit;
930 1890
931 default: 1891 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1892
938 X_LOCK (reslock); 1893 X_LOCK (reslock);
939 1894
940 ++npending; 1895 ++npending;
941 1896
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1897 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1898 want_poll_cb ();
944 1899
945 self->req = 0; 1900 self->req = 0;
946 worker_clear (self); 1901 etp_worker_clear (self);
947 1902
948 X_UNLOCK (reslock); 1903 X_UNLOCK (reslock);
949 } 1904 }
950 1905
951quit: 1906quit:
952 X_LOCK (wrklock); 1907 X_LOCK (wrklock);
953 worker_free (self); 1908 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1909 X_UNLOCK (wrklock);
955 1910
956 return 0; 1911 return 0;
957} 1912}
958 1913
959/*****************************************************************************/ 1914/*****************************************************************************/
960 1915
961static void eio_atfork_prepare (void) 1916int 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)) 1917eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1918{
1019 want_poll_cb = want_poll; 1919 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} 1920}
1035 1921
1922ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1923eio_api_destroy (eio_req *req)
1037{ 1924{
1038 free (req); 1925 free (req);
1039} 1926}
1040 1927
1041#define REQ(rtype) \ 1928#define REQ(rtype) \
1043 \ 1930 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1931 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1932 if (!req) \
1046 return 0; \ 1933 return 0; \
1047 \ 1934 \
1048 req->type = EIO_ ## rtype; \ 1935 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1936 req->pri = pri; \
1050 req->finish = cb; \ 1937 req->finish = cb; \
1938 req->data = data; \
1051 req->destroy = eio_api_destroy; 1939 req->destroy = eio_api_destroy;
1052 1940
1053#define SEND eio_submit (req); return req 1941#define SEND eio_submit (req); return req
1054 1942
1055#define PATH (void)0 1943#define PATH \
1944 req->flags |= EIO_FLAG_PTR1_FREE; \
1945 req->ptr1 = strdup (path); \
1946 if (!req->ptr1) \
1947 { \
1948 eio_api_destroy (req); \
1949 return 0; \
1950 }
1056 1951
1952static void
1953eio_execute (etp_worker *self, eio_req *req)
1954{
1955 if (ecb_expect_false (EIO_CANCELLED (req)))
1956 {
1957 req->result = -1;
1958 req->errorno = ECANCELED;
1959 return;
1960 }
1961
1962 switch (req->type)
1963 {
1964 case EIO_READ: ALLOC (req->size);
1965 req->result = req->offs >= 0
1966 ? pread (req->int1, req->ptr2, req->size, req->offs)
1967 : read (req->int1, req->ptr2, req->size); break;
1968 case EIO_WRITE: req->result = req->offs >= 0
1969 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1970 : write (req->int1, req->ptr2, req->size); break;
1971
1972 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1973 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1974
1975 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1976 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1977 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1979 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1980 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1981
1982 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1983 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1984 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986
1987 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1988 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1989 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1990 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1991 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1992 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1993
1994 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1995 case EIO_CLOSE: req->result = close (req->int1); break;
1996 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1997 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1998 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1999 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2000 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2001 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2002 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2003 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2004
2005 case EIO_REALPATH: eio__realpath (req, self); break;
2006
2007 case EIO_READLINK: ALLOC (PATH_MAX);
2008 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2009
2010 case EIO_SYNC: req->result = 0; sync (); break;
2011 case EIO_FSYNC: req->result = fsync (req->int1); break;
2012 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2013 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2014 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2015 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2016 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2017 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2018 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2019
2020 case EIO_READDIR: eio__scandir (req, self); break;
2021
2022 case EIO_BUSY:
2023#ifdef _WIN32
2024 Sleep (req->nv1 * 1e3);
2025#else
2026 {
2027 struct timeval tv;
2028
2029 tv.tv_sec = req->nv1;
2030 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2031
2032 req->result = select (0, 0, 0, 0, &tv);
2033 }
2034#endif
2035 break;
2036
2037 case EIO_UTIME:
2038 case EIO_FUTIME:
2039 {
2040 struct timeval tv[2];
2041 struct timeval *times;
2042
2043 if (req->nv1 != -1. || req->nv2 != -1.)
2044 {
2045 tv[0].tv_sec = req->nv1;
2046 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2047 tv[1].tv_sec = req->nv2;
2048 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2049
2050 times = tv;
2051 }
2052 else
2053 times = 0;
2054
2055 req->result = req->type == EIO_FUTIME
2056 ? futimes (req->int1, times)
2057 : utimes (req->ptr1, times);
2058 }
2059 break;
2060
2061 case EIO_GROUP:
2062 abort (); /* handled in eio_request */
2063
2064 case EIO_NOP:
2065 req->result = 0;
2066 break;
2067
2068 case EIO_CUSTOM:
2069 req->feed (req);
2070 break;
2071
2072 default:
2073 errno = ENOSYS;
2074 req->result = -1;
2075 break;
2076 }
2077
2078 req->errorno = errno;
2079}
2080
2081#ifndef EIO_NO_WRAPPERS
2082
2083eio_req *eio_nop (int pri, eio_cb cb, void *data)
2084{
2085 REQ (EIO_NOP); SEND;
2086}
2087
2088eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2089{
2090 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2091}
2092
1057eio_req *eio_fsync (int fd, eio_cb cb) 2093eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2094{
2095 REQ (EIO_SYNC); SEND;
2096}
2097
2098eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2099{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2100 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2101}
1061 2102
2103eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2104{
2105 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2106}
2107
2108eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2109{
2110 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2111}
2112
2113eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2114{
2115 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2116}
2117
2118eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2119{
2120 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2121}
2122
2123eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2124{
2125 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2126}
2127
2128eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2129{
2130 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2131}
2132
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2133eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2134{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2135 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2136}
1066 2137
2138eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2139{
2140 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2141}
2142
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2143eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2144{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2145 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2146}
1071 2147
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2148eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2149{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2150 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2151}
1076 2152
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2153eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2154{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2155 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2156}
1081 2157
1082eio_req *eio_fstat (int fd, eio_cb cb) 2158eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2159{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2160 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2161}
1086 2162
2163eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2164{
2165 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2166}
2167
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2168eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2169{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2170 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2171}
1091 2172
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2173eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2174{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2175 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2176}
1096 2177
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2178eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2179{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2180 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2181}
1101 2182
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2183eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2184{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2185 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2186}
1106 2187
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2188eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2189{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2190 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2191}
1111 2192
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2193eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2194{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2195 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2196}
1116 2197
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2198eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2199{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2200 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2201}
1121 2202
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) 2203eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2204{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2205 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2206}
1141 2207
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2208eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2209{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2210 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2211}
1146 2212
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2213eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2214{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2215 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2216}
1151 2217
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2218eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2219{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2220 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2221}
1156 2222
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2223eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2224{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2225 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2226}
1161 2227
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2228static eio_req *
2229eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2230{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2231 REQ (type); PATH; SEND;
1170} 2232}
1171 2233
2234eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2235{
2236 return eio__1path (EIO_READLINK, path, pri, cb, data);
2237}
2238
2239eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2240{
2241 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2242}
2243
2244eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2245{
2246 return eio__1path (EIO_STAT, path, pri, cb, data);
2247}
2248
2249eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2250{
2251 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2252}
2253
2254eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2255{
2256 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2257}
2258
2259eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2260{
2261 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2262}
2263
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2264eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2265{
1174 REQ (READDIR); PATH; SEND; 2266 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2267}
1176 2268
2269eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2270{
2271 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2272}
2273
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2274eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2275{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2276 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2277}
1181 2278
1182eio_req *eio_busy (double delay, eio_cb cb) 2279static eio_req *
2280eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2281{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2282 REQ (type); PATH;
1185}
1186 2283
2284 req->flags |= EIO_FLAG_PTR2_FREE;
2285 req->ptr2 = strdup (new_path);
2286 if (!req->ptr2)
2287 {
2288 eio_api_destroy (req);
2289 return 0;
2290 }
2291
2292 SEND;
2293}
2294
2295eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2296{
2297 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2298}
2299
2300eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2301{
2302 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2303}
2304
2305eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2306{
2307 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2308}
2309
2310eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2311{
2312 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2313}
2314
2315#endif
2316
1187eio_req *eio_nop (eio_cb cb) 2317eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2318{
2319 const int pri = EIO_PRI_MAX;
2320
1189 REQ (NOP); SEND; 2321 REQ (EIO_GROUP); SEND;
1190} 2322}
1191 2323
1192#undef REQ 2324#undef REQ
1193#undef PATH 2325#undef PATH
1194#undef SEND 2326#undef SEND
1195 2327
1196#if 0 2328/*****************************************************************************/
2329/* grp functions */
1197 2330
1198void 2331void
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) 2332eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2333{
1573 grp->int2 = limit; 2334 grp->int2 = limit;
1574 grp->feed = feed; 2335 grp->feed = feed;
1575 2336
1576 grp_try_feed (grp); 2337 grp_try_feed (grp);
1577} 2338}
1578 2339
2340void
1579void eio_grp_limit (eio_req *grp, int limit) 2341eio_grp_limit (eio_req *grp, int limit)
1580{ 2342{
1581 grp->int2 = limit; 2343 grp->int2 = limit;
1582 2344
1583 grp_try_feed (grp); 2345 grp_try_feed (grp);
1584} 2346}
1585 2347
2348void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2349eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2350{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2351 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2352
2353 grp->flags |= EIO_FLAG_GROUPADD;
1589 2354
1590 ++grp->size; 2355 ++grp->size;
1591 req->grp = grp; 2356 req->grp = grp;
1592 2357
1593 req->grp_prev = 0; 2358 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2362 grp->grp_first->grp_prev = req;
1598 2363
1599 grp->grp_first = req; 2364 grp->grp_first = req;
1600} 2365}
1601 2366
2367/*****************************************************************************/
2368/* misc garbage */
1602 2369
2370ssize_t
2371eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2372{
2373 etp_worker wrk;
2374 ssize_t ret;
2375
2376 wrk.dbuf = 0;
2377
2378 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2379
2380 if (wrk.dbuf)
2381 free (wrk.dbuf);
2382
2383 return ret;
2384}
2385

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines