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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.43 by root, Wed Jun 17 18:29:15 2009 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009 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
1#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
2#include "xthread.h" 45#include "xthread.h"
3 46
4#include <errno.h> 47#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 48#include <stddef.h>
11#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h>
12#include <errno.h> 51#include <errno.h>
13#include <sys/types.h> 52#include <sys/types.h>
14#include <sys/stat.h> 53#include <sys/stat.h>
15#include <limits.h> 54#include <limits.h>
16#include <fcntl.h> 55#include <fcntl.h>
17#include <sched.h> 56#include <assert.h>
18 57
19#ifndef EIO_FINISH 58#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 60#endif
22 61
29#endif 68#endif
30 69
31#ifdef _WIN32 70#ifdef _WIN32
32 71
33 /*doh*/ 72 /*doh*/
34
35#else 73#else
36 74
37# include "config.h" 75# include "config.h"
38# include <sys/time.h> 76# include <sys/time.h>
39# include <sys/select.h> 77# include <sys/select.h>
78# include <sys/mman.h>
40# include <unistd.h> 79# include <unistd.h>
41# include <utime.h> 80# include <utime.h>
42# include <signal.h> 81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif
92
93#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif
43 101
44# ifndef EIO_STRUCT_DIRENT 102# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 103# define EIO_STRUCT_DIRENT struct dirent
46# endif 104# endif
47 105
48#endif 106#endif
49
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53 107
54#if HAVE_SENDFILE 108#if HAVE_SENDFILE
55# if __linux 109# if __linux
56# include <sys/sendfile.h> 110# include <sys/sendfile.h>
57# elif __freebsd 111# elif __freebsd
64# else 118# else
65# error sendfile support requested but not available 119# error sendfile support requested but not available
66# endif 120# endif
67#endif 121#endif
68 122
123#ifndef D_TYPE
124# define D_TYPE(de) 0
125#endif
126#ifndef D_INO
127# define D_INO(de) 0
128#endif
129#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif
132
69/* number of seconds after which an idle threads exit */ 133/* number of seconds after which an idle threads exit */
70#define IDLE_TIMEOUT 10 134#define IDLE_TIMEOUT 10
71 135
72/* used for struct dirent, AIX doesn't provide it */ 136/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 137#ifndef NAME_MAX
77/* buffer size for various temporary buffers */ 141/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 142#define EIO_BUFSIZE 65536
79 143
80#define dBUF \ 144#define dBUF \
81 char *eio_buf; \ 145 char *eio_buf; \
82 X_LOCK (wrklock); \ 146 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \
85 if (!eio_buf) \ 150 if (!eio_buf) \
86 return -1; 151 return -1;
87 152
88#define EIO_TICKS ((1000000 + 1023) >> 10) 153#define EIO_TICKS ((1000000 + 1023) >> 10)
89 154
155/*****************************************************************************/
156
157#if __GNUC__ >= 3
158# define expect(expr,value) __builtin_expect ((expr),(value))
159#else
160# define expect(expr,value) (expr)
161#endif
162
163#define expect_false(expr) expect ((expr) != 0, 0)
164#define expect_true(expr) expect ((expr) != 0, 1)
165
166/*****************************************************************************/
167
168#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX
170
171struct etp_worker;
172
173#define ETP_REQ eio_req
174#define ETP_DESTROY(req) eio_destroy (req)
175static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179
180#define ETP_WORKER_CLEAR(req) \
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192
193#define ETP_WORKER_COMMON \
194 void *dbuf; \
195 DIR *dirp;
196
197/*****************************************************************************/
198
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200
201/* calculcate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206}
207
208static unsigned int started, idle, wanted = 4;
209
90static void (*want_poll_cb) (void); 210static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 211static void (*done_poll_cb) (void);
92 212
93static unsigned int max_poll_time = 0; 213static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 214static unsigned int max_poll_reqs; /* reslock */
95 215
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 216static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static volatile unsigned int nready; /* reqlock */
98{ 218static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219static volatile unsigned int max_idle = 4;
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
101}
102 220
103static unsigned int started, idle, wanted;
104
105/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT; 221static mutex_t wrklock = X_MUTEX_INIT;
107
108typedef struct worker {
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk)
123{
124 if (wrk->dirp)
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135}
136
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next;
141
142 free (wrk);
143}
144
145static volatile unsigned int nreqs, nready, npending;
146static volatile unsigned int max_idle = 4;
147
148static mutex_t reslock = X_MUTEX_INIT; 222static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT; 223static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT; 224static cond_t reqwait = X_COND_INIT;
151 225
226#if !HAVE_PREADWRITE
227/*
228 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine.
231 */
232static mutex_t preadwritelock = X_MUTEX_INIT;
233#endif
234
235typedef struct etp_worker
236{
237 /* locked by wrklock */
238 struct etp_worker *prev, *next;
239
240 thread_t tid;
241
242 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON
246} etp_worker;
247
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
249
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252
253/* worker threads management */
254
255static void etp_worker_clear (etp_worker *wrk)
256{
257 ETP_WORKER_CLEAR (wrk);
258}
259
260static void etp_worker_free (etp_worker *wrk)
261{
262 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next;
264
265 free (wrk);
266}
267
152unsigned int eio_nreqs (void) 268static unsigned int etp_nreqs (void)
153{ 269{
270 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 274 return retval;
155} 275}
156 276
157unsigned int eio_nready (void) 277static unsigned int etp_nready (void)
158{ 278{
159 unsigned int retval; 279 unsigned int retval;
160 280
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 282 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 284
165 return retval; 285 return retval;
166} 286}
167 287
168unsigned int eio_npending (void) 288static unsigned int etp_npending (void)
169{ 289{
170 unsigned int retval; 290 unsigned int retval;
171 291
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 293 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 295
176 return retval; 296 return retval;
177} 297}
178 298
179unsigned int eio_nthreads (void) 299static unsigned int etp_nthreads (void)
180{ 300{
181 unsigned int retval; 301 unsigned int retval;
182 302
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 304 retval = started;
191 * a somewhat faster data structure might be nice, but 311 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 312 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 313 * per shift, the most expensive operation.
194 */ 314 */
195typedef struct { 315typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 316 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 317 int size;
198} reqq; 318} etp_reqq;
199 319
200static reqq req_queue; 320static etp_reqq req_queue;
201static reqq res_queue; 321static etp_reqq res_queue;
202 322
203static int reqq_push (reqq *q, eio_req *req) 323static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 324{
205 int pri = req->pri; 325 int pri = req->pri;
206 req->next = 0; 326 req->next = 0;
207 327
208 if (q->qe[pri]) 328 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 334 q->qe[pri] = q->qs[pri] = req;
215 335
216 return q->size++; 336 return q->size++;
217} 337}
218 338
219static eio_req *reqq_shift (reqq *q) 339static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 340{
221 int pri; 341 int pri;
222 342
223 if (!q->size) 343 if (!q->size)
224 return 0; 344 return 0;
225 345
226 --q->size; 346 --q->size;
227 347
228 for (pri = EIO_NUM_PRI; pri--; ) 348 for (pri = ETP_NUM_PRI; pri--; )
229 { 349 {
230 eio_req *req = q->qs[pri]; 350 eio_req *req = q->qs[pri];
231 351
232 if (req) 352 if (req)
233 { 353 {
239 } 359 }
240 360
241 abort (); 361 abort ();
242} 362}
243 363
244static void grp_try_feed (eio_req *grp) 364static void etp_atfork_prepare (void)
245{ 365{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 366 X_LOCK (wrklock);
367 X_LOCK (reqlock);
368 X_LOCK (reslock);
369#if !HAVE_PREADWRITE
370 X_LOCK (preadwritelock);
371#endif
372}
373
374static void etp_atfork_parent (void)
375{
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383
384static void etp_atfork_child (void)
385{
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
247 { 395 {
248 int old_len = grp->size; 396 etp_worker *wrk = wrk_first.next;
249 397
250 EIO_FEED (grp); 398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
251 400
252 /* stop if no progress has been made */ 401 etp_worker_clear (wrk);
253 if (old_len == grp->size) 402 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 403 }
259}
260 404
261static int eio_finish (eio_req *req); 405 started = 0;
406 idle = 0;
407 nreqs = 0;
408 nready = 0;
409 npending = 0;
262 410
263static int grp_dec (eio_req *grp) 411 etp_atfork_parent ();
264{ 412}
265 --grp->size;
266 413
267 /* call feeder, if applicable */ 414static void
268 grp_try_feed (grp); 415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
269 419
270 /* finish, if done */ 420static int
271 if (!grp->size && grp->int1) 421etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 422{
273 else 423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll;
428 done_poll_cb = done_poll;
429
274 return 0; 430 return 0;
275} 431}
276 432
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 433X_THREAD_PROC (etp_proc);
326 434
327static void start_thread (void) 435static void etp_start_thread (void)
328{ 436{
329 worker *wrk = calloc (1, sizeof (worker)); 437 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 438
331 if (!wrk) 439 /*TODO*/
332 croak ("unable to allocate worker thread data"); 440 assert (("unable to allocate worker thread data", wrk));
333 441
334 X_LOCK (wrklock); 442 X_LOCK (wrklock);
335 443
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 444 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 445 {
338 wrk->prev = &wrk_first; 446 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 447 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 448 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 449 wrk_first.next = wrk;
345 free (wrk); 453 free (wrk);
346 454
347 X_UNLOCK (wrklock); 455 X_UNLOCK (wrklock);
348} 456}
349 457
350static void maybe_start_thread (void) 458static void etp_maybe_start_thread (void)
351{ 459{
352 if (eio_nthreads () >= wanted) 460 if (expect_true (etp_nthreads () >= wanted))
353 return; 461 return;
354 462
355 /* todo: maybe use idle here, but might be less exact */ 463 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 465 return;
358 466
359 start_thread (); 467 etp_start_thread ();
360} 468}
361 469
362void eio_submit (eio_req *req)
363{
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void) 470static void etp_end_thread (void)
376{ 471{
377 eio_req *req = calloc (1, sizeof (eio_req)); 472 eio_req *req = calloc (1, sizeof (eio_req));
378 473
379 req->type = EIO_QUIT; 474 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 476
382 X_LOCK (reqlock); 477 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 478 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 479 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 480 X_UNLOCK (reqlock);
387 X_LOCK (wrklock); 482 X_LOCK (wrklock);
388 --started; 483 --started;
389 X_UNLOCK (wrklock); 484 X_UNLOCK (wrklock);
390} 485}
391 486
392void eio_set_max_poll_time (double nseconds)
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 487static int etp_poll (void)
429{ 488{
430 int maxreqs = max_poll_reqs; 489 unsigned int maxreqs;
490 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 491 struct timeval tv_start, tv_now;
432 eio_req *req;
433 492
493 X_LOCK (reslock);
494 maxreqs = max_poll_reqs;
495 maxtime = max_poll_time;
496 X_UNLOCK (reslock);
497
434 if (max_poll_time) 498 if (maxtime)
435 gettimeofday (&tv_start, 0); 499 gettimeofday (&tv_start, 0);
436 500
437 for (;;) 501 for (;;)
438 { 502 {
503 ETP_REQ *req;
504
439 maybe_start_thread (); 505 etp_maybe_start_thread ();
440 506
441 X_LOCK (reslock); 507 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 508 req = reqq_shift (&res_queue);
443 509
444 if (req) 510 if (req)
452 X_UNLOCK (reslock); 518 X_UNLOCK (reslock);
453 519
454 if (!req) 520 if (!req)
455 return 0; 521 return 0;
456 522
523 X_LOCK (reqlock);
457 --nreqs; 524 --nreqs;
525 X_UNLOCK (reqlock);
458 526
459 if (req->type == EIO_GROUP && req->size) 527 if (expect_false (req->type == EIO_GROUP && req->size))
460 { 528 {
461 req->int1 = 1; /* mark request as delayed */ 529 req->int1 = 1; /* mark request as delayed */
462 continue; 530 continue;
463 } 531 }
464 else 532 else
465 { 533 {
466 int res = eio_finish (req); 534 int res = ETP_FINISH (req);
467 if (res) 535 if (expect_false (res))
468 return res; 536 return res;
469 } 537 }
470 538
471 if (maxreqs && !--maxreqs) 539 if (expect_false (maxreqs && !--maxreqs))
472 break; 540 break;
473 541
474 if (max_poll_time) 542 if (maxtime)
475 { 543 {
476 gettimeofday (&tv_now, 0); 544 gettimeofday (&tv_now, 0);
477 545
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 546 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 547 break;
480 } 548 }
481 } 549 }
482 550
483 errno = EAGAIN; 551 errno = EAGAIN;
484 return -1; 552 return -1;
485} 553}
486 554
555static void etp_cancel (ETP_REQ *req)
556{
557 X_LOCK (wrklock);
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560
561 eio_grp_cancel (req);
562}
563
564static void etp_submit (ETP_REQ *req)
565{
566 req->pri -= ETP_PRI_MIN;
567
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
574 X_LOCK (reqlock);
575 ++nreqs;
576 X_UNLOCK (reqlock);
577
578 X_LOCK (reslock);
579
580 ++npending;
581
582 if (!reqq_push (&res_queue, req) && want_poll_cb)
583 want_poll_cb ();
584
585 X_UNLOCK (reslock);
586 }
587 else
588 {
589 X_LOCK (reqlock);
590 ++nreqs;
591 ++nready;
592 reqq_push (&req_queue, req);
593 X_COND_SIGNAL (reqwait);
594 X_UNLOCK (reqlock);
595
596 etp_maybe_start_thread ();
597 }
598}
599
600static void etp_set_max_poll_time (double nseconds)
601{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605}
606
607static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612}
613
614static void etp_set_max_idle (unsigned int nthreads)
615{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619}
620
621static void etp_set_min_parallel (unsigned int nthreads)
622{
623 if (wanted < nthreads)
624 wanted = nthreads;
625}
626
627static void etp_set_max_parallel (unsigned int nthreads)
628{
629 if (wanted > nthreads)
630 wanted = nthreads;
631
632 while (started > wanted)
633 etp_end_thread ();
634}
635
636/*****************************************************************************/
637
638static void grp_try_feed (eio_req *grp)
639{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 {
642 grp->flags &= ~EIO_FLAG_GROUPADD;
643
644 EIO_FEED (grp);
645
646 /* stop if no progress has been made */
647 if (!(grp->flags & EIO_FLAG_GROUPADD))
648 {
649 grp->feed = 0;
650 break;
651 }
652 }
653}
654
655static int grp_dec (eio_req *grp)
656{
657 --grp->size;
658
659 /* call feeder, if applicable */
660 grp_try_feed (grp);
661
662 /* finish, if done */
663 if (!grp->size && grp->int1)
664 return eio_finish (grp);
665 else
666 return 0;
667}
668
669void eio_destroy (eio_req *req)
670{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673
674 EIO_DESTROY (req);
675}
676
677static int eio_finish (eio_req *req)
678{
679 int res = EIO_FINISH (req);
680
681 if (req->grp)
682 {
683 int res2;
684 eio_req *grp = req->grp;
685
686 /* unlink request */
687 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
688 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
689
690 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next;
692
693 res2 = grp_dec (grp);
694
695 if (!res && res2)
696 res = res2;
697 }
698
699 eio_destroy (req);
700
701 return res;
702}
703
704void eio_grp_cancel (eio_req *grp)
705{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp);
708}
709
710void eio_cancel (eio_req *req)
711{
712 etp_cancel (req);
713}
714
715void eio_submit (eio_req *req)
716{
717 etp_submit (req);
718}
719
720unsigned int eio_nreqs (void)
721{
722 return etp_nreqs ();
723}
724
725unsigned int eio_nready (void)
726{
727 return etp_nready ();
728}
729
730unsigned int eio_npending (void)
731{
732 return etp_npending ();
733}
734
735unsigned int eio_nthreads (void)
736{
737 return etp_nthreads ();
738}
739
740void eio_set_max_poll_time (double nseconds)
741{
742 etp_set_max_poll_time (nseconds);
743}
744
745void eio_set_max_poll_reqs (unsigned int maxreqs)
746{
747 etp_set_max_poll_reqs (maxreqs);
748}
749
750void eio_set_max_idle (unsigned int nthreads)
751{
752 etp_set_max_idle (nthreads);
753}
754
755void eio_set_min_parallel (unsigned int nthreads)
756{
757 etp_set_min_parallel (nthreads);
758}
759
760void eio_set_max_parallel (unsigned int nthreads)
761{
762 etp_set_max_parallel (nthreads);
763}
764
765int eio_poll (void)
766{
767 return etp_poll ();
768}
769
487/*****************************************************************************/ 770/*****************************************************************************/
488/* work around various missing functions */ 771/* work around various missing functions */
489 772
490#if !HAVE_PREADWRITE 773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
491# define pread aio_pread 776# define pread eio__pread
492# define pwrite aio_pwrite 777# define pwrite eio__pwrite
493 778
494/* 779static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 780eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 781{
503 ssize_t res; 782 ssize_t res;
504 off_t ooffset; 783 off_t ooffset;
505 784
506 X_LOCK (preadwritelock); 785 X_LOCK (preadwritelock);
511 X_UNLOCK (preadwritelock); 790 X_UNLOCK (preadwritelock);
512 791
513 return res; 792 return res;
514} 793}
515 794
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 795static ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 797{
518 ssize_t res; 798 ssize_t res;
519 off_t ooffset; 799 off_t ooffset;
520 800
521 X_LOCK (preadwritelock); 801 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 802 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 803 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 804 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 805 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 806 X_UNLOCK (preadwritelock);
527 807
528 return res; 808 return res;
529} 809}
530#endif 810#endif
531 811
532#ifndef HAVE_FUTIMES 812#ifndef HAVE_FUTIMES
533 813
814# undef utimes
815# undef futimes
534# define utimes(path,times) aio_utimes (path, times) 816# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times) 817# define futimes(fd,times) eio__futimes (fd, times)
536 818
819static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 820eio__utimes (const char *filename, const struct timeval times[2])
538{ 821{
539 if (times) 822 if (times)
540 { 823 {
541 struct utimbuf buf; 824 struct utimbuf buf;
542 825
547 } 830 }
548 else 831 else
549 return utime (filename, 0); 832 return utime (filename, 0);
550} 833}
551 834
552static int aio_futimes (int fd, const struct timeval tv[2]) 835static int eio__futimes (int fd, const struct timeval tv[2])
553{ 836{
554 errno = ENOSYS; 837 errno = ENOSYS;
555 return -1; 838 return -1;
556} 839}
557 840
558#endif 841#endif
559 842
560#if !HAVE_FDATASYNC 843#if !HAVE_FDATASYNC
844# undef fdatasync
561# define fdatasync fsync 845# define fdatasync(fd) fsync (fd)
562#endif 846#endif
847
848/* sync_file_range always needs emulation */
849int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{
852#if HAVE_SYNC_FILE_RANGE
853 int res;
854
855 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
856 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
857 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
858 {
859 flags = 0
860 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
861 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
862 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
863 }
864
865 res = sync_file_range (fd, offset, nbytes, flags);
866
867 if (!res || errno != ENOSYS)
868 return res;
869#endif
870
871 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd);
874}
563 875
564#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
877# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 879
880static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 882{
569 size_t todo = count; 883 size_t todo = count;
570 dBUF; 884 dBUF;
571 885
572 while (todo > 0) 886 while (todo > 0)
573 { 887 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 888 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 889
576 pread (fd, aio_buf, len, offset); 890 pread (fd, eio_buf, len, offset);
577 offset += len; 891 offset += len;
578 todo -= len; 892 todo -= len;
579 } 893 }
580 894
581 errno = 0; 895 errno = 0;
582 return count; 896 return count;
583} 897}
584 898
585#endif 899#endif
586 900
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 901/* sendfile always needs emulation */
902static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 904{
620 ssize_t res; 905 ssize_t res;
621 906
622 if (!count) 907 if (!count)
623 return 0; 908 return 0;
705 } 990 }
706 991
707 return res; 992 return res;
708} 993}
709 994
995static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{
998 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1000}
1001
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003
1004#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1005#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1006
1007static void
1008eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1009{
1010 unsigned char bits [9 + sizeof (ino_t) * 8];
1011 unsigned char *bit = bits;
1012
1013 assert (CHAR_BIT == 8);
1014 assert (sizeof (eio_dirent) * 8 < 256);
1015 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1016 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1017
1018 if (size <= EIO_SORT_FAST)
1019 return;
1020
1021 /* first prepare an array of bits to test in our radix sort */
1022 /* try to take endianness into account, as well as differences in ino_t sizes */
1023 /* inode_bits must contain all inodes ORed together */
1024 /* which is used to skip bits that are 0 everywhere, which is very common */
1025 {
1026 ino_t endianness;
1027 int i, j;
1028
1029 /* we store the byte offset of byte n into byte n of "endianness" */
1030 for (i = 0; i < sizeof (ino_t); ++i)
1031 ((unsigned char *)&endianness)[i] = i;
1032
1033 *bit++ = 0;
1034
1035 for (i = 0; i < sizeof (ino_t); ++i)
1036 {
1037 /* shifting off the byte offsets out of "endianness" */
1038 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1039 endianness >>= 8;
1040
1041 for (j = 0; j < 8; ++j)
1042 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1043 *bit++ = offs + j;
1044 }
1045
1046 for (j = 0; j < 8; ++j)
1047 if (score_bits & (1 << j))
1048 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1049 }
1050
1051 /* now actually do the sorting (a variant of MSD radix sort) */
1052 {
1053 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1054 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1055 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1056 int stk_idx = 0;
1057
1058 base_stk [stk_idx] = dents;
1059 end_stk [stk_idx] = dents + size;
1060 bit_stk [stk_idx] = bit - 1;
1061
1062 do
1063 {
1064 base = base_stk [stk_idx];
1065 end = end_stk [stk_idx];
1066 bit = bit_stk [stk_idx];
1067
1068 for (;;)
1069 {
1070 unsigned char O = *bit >> 3;
1071 unsigned char M = 1 << (*bit & 7);
1072
1073 eio_dirent *a = base;
1074 eio_dirent *b = end;
1075
1076 if (b - a < EIO_SORT_CUTOFF)
1077 break;
1078
1079 /* now bit-partition the array on the bit */
1080 /* this ugly asymmetric loop seems to perform much better than typical */
1081 /* partition algos found in the literature */
1082 do
1083 if (!(((unsigned char *)a)[O] & M))
1084 ++a;
1085 else if (!(((unsigned char *)--b)[O] & M))
1086 {
1087 eio_dirent tmp = *a; *a = *b; *b = tmp;
1088 ++a;
1089 }
1090 while (b > a);
1091
1092 /* next bit, or stop, if no bits left in this path */
1093 if (!*--bit)
1094 break;
1095
1096 base_stk [stk_idx] = a;
1097 end_stk [stk_idx] = end;
1098 bit_stk [stk_idx] = bit;
1099 ++stk_idx;
1100
1101 end = a;
1102 }
1103 }
1104 while (stk_idx--);
1105 }
1106}
1107
1108static void
1109eio_dent_insertion_sort (eio_dirent *dents, int size)
1110{
1111 /* first move the smallest element to the front, to act as a sentinel */
1112 {
1113 int i;
1114 eio_dirent *min = dents;
1115
1116 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1117 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1118 if (EIO_DENT_CMP (dents [i], <, *min))
1119 min = &dents [i];
1120
1121 /* swap elements 0 and j (minimum) */
1122 {
1123 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1124 }
1125 }
1126
1127 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1128 {
1129 eio_dirent *i, *j;
1130
1131 for (i = dents + 1; i < dents + size; ++i)
1132 {
1133 eio_dirent value = *i;
1134
1135 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1136 j [1] = j [0];
1137
1138 j [1] = value;
1139 }
1140 }
1141}
1142
1143static void
1144eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1145{
1146 if (size <= 1)
1147 return; /* our insertion sort relies on size > 0 */
1148
1149 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1150 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1151 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1152
1153 /* use an insertion sort at the end, or for small arrays, */
1154 /* as insertion sort is more efficient for small partitions */
1155 eio_dent_insertion_sort (dents, size);
1156}
1157
710/* read a full directory */ 1158/* read a full directory */
1159static void
711static void scandir_ (eio_req *req, worker *self) 1160eio__scandir (eio_req *req, etp_worker *self)
712{ 1161{
713 DIR *dirp; 1162 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1163 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1164 char *name, *names;
721 int memlen = 4096; 1165 int namesalloc = 4096;
722 int memofs = 0; 1166 int namesoffs = 0;
1167 int flags = req->int1;
1168 eio_dirent *dents = 0;
1169 int dentalloc = 128;
723 int res = 0; 1170 int dentoffs = 0;
1171 ino_t inode_bits = 0;
1172
1173 req->result = -1;
1174
1175 if (!(flags & EIO_READDIR_DENTS))
1176 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1177
725 X_LOCK (wrklock); 1178 X_LOCK (wrklock);
1179 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1180 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE; 1181 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1182 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1183 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1184 X_UNLOCK (wrklock);
731 1185
732 if (dirp && u && names) 1186 if (dirp && names && (!flags || dents))
733 for (;;) 1187 for (;;)
734 { 1188 {
735 errno = 0; 1189 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1190 entp = readdir (dirp);
737 1191
738 if (!entp) 1192 if (!entp)
1193 {
1194 if (errno)
1195 break;
1196
1197 /* sort etc. */
1198 req->int1 = flags;
1199 req->result = dentoffs;
1200
1201 if (flags & EIO_READDIR_STAT_ORDER)
1202 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1203 else if (flags & EIO_READDIR_DIRS_FIRST)
1204 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1205 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1206 else
1207 {
1208 /* in this case, all is known, and we just put dirs first and sort them */
1209 eio_dirent *oth = dents + dentoffs;
1210 eio_dirent *dir = dents;
1211
1212 /* now partition dirs to the front, and non-dirs to the back */
1213 /* by walking from both sides and swapping if necessary */
1214 /* also clear score, so it doesn't influence sorting */
1215 while (oth > dir)
1216 {
1217 if (dir->type == EIO_DT_DIR)
1218 ++dir;
1219 else if ((--oth)->type == EIO_DT_DIR)
1220 {
1221 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1222
1223 ++dir;
1224 }
1225 }
1226
1227 /* now sort the dirs only */
1228 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1229 }
1230
739 break; 1231 break;
1232 }
740 1233
1234 /* now add the entry to our list(s) */
741 name = entp->d_name; 1235 name = entp->d_name;
742 1236
1237 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1238 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1239 {
745 int len = strlen (name) + 1; 1240 int len = D_NAMLEN (entp) + 1;
746 1241
747 res++; 1242 while (expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1243 {
751 memlen *= 2; 1244 namesalloc *= 2;
752 X_LOCK (wrklock); 1245 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1246 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1247 X_UNLOCK (wrklock);
755 1248
756 if (!names) 1249 if (!names)
757 break; 1250 break;
758 } 1251 }
759 1252
760 memcpy (names + memofs, name, len); 1253 memcpy (names + namesoffs, name, len);
1254
1255 if (dents)
1256 {
1257 struct eio_dirent *ent;
1258
1259 if (expect_false (dentoffs == dentalloc))
1260 {
1261 dentalloc *= 2;
1262 X_LOCK (wrklock);
1263 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1264 X_UNLOCK (wrklock);
1265
1266 if (!dents)
1267 break;
1268 }
1269
1270 ent = dents + dentoffs;
1271
1272 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1273 ent->namelen = len - 1;
1274 ent->inode = D_INO (entp);
1275
1276 inode_bits |= ent->inode;
1277
1278 switch (D_TYPE (entp))
1279 {
1280 default:
1281 ent->type = EIO_DT_UNKNOWN;
1282 flags |= EIO_READDIR_FOUND_UNKNOWN;
1283 break;
1284
1285 #ifdef DT_FIFO
1286 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1287 #endif
1288 #ifdef DT_CHR
1289 case DT_CHR: ent->type = EIO_DT_CHR; break;
1290 #endif
1291 #ifdef DT_MPC
1292 case DT_MPC: ent->type = EIO_DT_MPC; break;
1293 #endif
1294 #ifdef DT_DIR
1295 case DT_DIR: ent->type = EIO_DT_DIR; break;
1296 #endif
1297 #ifdef DT_NAM
1298 case DT_NAM: ent->type = EIO_DT_NAM; break;
1299 #endif
1300 #ifdef DT_BLK
1301 case DT_BLK: ent->type = EIO_DT_BLK; break;
1302 #endif
1303 #ifdef DT_MPB
1304 case DT_MPB: ent->type = EIO_DT_MPB; break;
1305 #endif
1306 #ifdef DT_REG
1307 case DT_REG: ent->type = EIO_DT_REG; break;
1308 #endif
1309 #ifdef DT_NWK
1310 case DT_NWK: ent->type = EIO_DT_NWK; break;
1311 #endif
1312 #ifdef DT_CMP
1313 case DT_CMP: ent->type = EIO_DT_CMP; break;
1314 #endif
1315 #ifdef DT_LNK
1316 case DT_LNK: ent->type = EIO_DT_LNK; break;
1317 #endif
1318 #ifdef DT_SOCK
1319 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1320 #endif
1321 #ifdef DT_DOOR
1322 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1323 #endif
1324 #ifdef DT_WHT
1325 case DT_WHT: ent->type = EIO_DT_WHT; break;
1326 #endif
1327 }
1328
1329 ent->score = 7;
1330
1331 if (flags & EIO_READDIR_DIRS_FIRST)
1332 {
1333 if (ent->type == EIO_DT_UNKNOWN)
1334 {
1335 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1336 ent->score = 1;
1337 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1338 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1339 }
1340 else if (ent->type == EIO_DT_DIR)
1341 ent->score = 0;
1342 }
1343 }
1344
761 memofs += len; 1345 namesoffs += len;
1346 ++dentoffs;
1347 }
1348
1349 if (EIO_CANCELLED (req))
1350 {
1351 errno = ECANCELED;
1352 break;
762 } 1353 }
763 } 1354 }
1355}
764 1356
765 if (errno) 1357#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
766 res = -1; 1358# undef msync
767 1359# define msync(a,b,c) ((errno = ENOSYS), -1)
768 req->result = res; 1360#endif
1361
1362int
1363eio__mtouch (void *mem, size_t len, int flags)
1364{
1365 intptr_t addr = (intptr_t)mem;
1366 intptr_t end = addr + len;
1367#ifdef PAGESIZE
1368 const intptr_t page = PAGESIZE;
1369#else
1370 static intptr_t page;
1371
1372 if (!page)
1373 page = sysconf (_SC_PAGESIZE);
1374#endif
1375
1376 addr &= ~(page - 1); /* assume page size is always a power of two */
1377
1378 if (addr < end)
1379 if (flags) /* modify */
1380 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1381 else
1382 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1383
1384 return 0;
769} 1385}
770 1386
771/*****************************************************************************/ 1387/*****************************************************************************/
772 1388
1389#define ALLOC(len) \
1390 if (!req->ptr2) \
1391 { \
1392 X_LOCK (wrklock); \
1393 req->flags |= EIO_FLAG_PTR2_FREE; \
1394 X_UNLOCK (wrklock); \
1395 req->ptr2 = malloc (len); \
1396 if (!req->ptr2) \
1397 { \
1398 errno = ENOMEM; \
1399 req->result = -1; \
1400 break; \
1401 } \
1402 }
1403
773X_THREAD_PROC (eio_proc) 1404X_THREAD_PROC (etp_proc)
774{ 1405{
775 eio_req *req; 1406 ETP_REQ *req;
776 struct timespec ts; 1407 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1408 etp_worker *self = (etp_worker *)thr_arg;
778 1409
779 /* try to distribute timeouts somewhat randomly */ 1410 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1411 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1412
782 for (;;) 1413 for (;;)
783 { 1414 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1415 X_LOCK (reqlock);
787 1416
788 for (;;) 1417 for (;;)
789 { 1418 {
790 self->req = req = reqq_shift (&req_queue); 1419 self->req = req = reqq_shift (&req_queue);
792 if (req) 1421 if (req)
793 break; 1422 break;
794 1423
795 ++idle; 1424 ++idle;
796 1425
1426 ts.tv_sec = time (0) + IDLE_TIMEOUT;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1427 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1428 {
799 if (idle > max_idle) 1429 if (idle > max_idle)
800 { 1430 {
801 --idle; 1431 --idle;
806 goto quit; 1436 goto quit;
807 } 1437 }
808 1438
809 /* we are allowed to idle, so do so without any timeout */ 1439 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1440 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1441 }
813 1442
814 --idle; 1443 --idle;
815 } 1444 }
816 1445
817 --nready; 1446 --nready;
818 1447
819 X_UNLOCK (reqlock); 1448 X_UNLOCK (reqlock);
820 1449
821 errno = 0; /* strictly unnecessary */ 1450 if (req->type < 0)
1451 goto quit;
822 1452
823 if (!EIO_CANCELLED (req)) 1453 if (!EIO_CANCELLED (req))
824 switch (req->type) 1454 ETP_EXECUTE (self, req);
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1455
918 X_LOCK (reslock); 1456 X_LOCK (reslock);
919 1457
920 ++npending; 1458 ++npending;
921 1459
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1460 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1461 want_poll_cb ();
924 1462
925 self->req = 0; 1463 self->req = 0;
926 worker_clear (self); 1464 etp_worker_clear (self);
927 1465
928 X_UNLOCK (reslock); 1466 X_UNLOCK (reslock);
929 } 1467 }
930 1468
931quit: 1469quit:
932 X_LOCK (wrklock); 1470 X_LOCK (wrklock);
933 worker_free (self); 1471 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1472 X_UNLOCK (wrklock);
935 1473
936 return 0; 1474 return 0;
937} 1475}
938 1476
939/*****************************************************************************/ 1477/*****************************************************************************/
940 1478
941static void eio_atfork_prepare (void) 1479int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1480{
943 X_LOCK (wrklock); 1481 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1482}
953 1483
954static void eio_atfork_parent (void) 1484static void eio_api_destroy (eio_req *req)
955{ 1485{
956#if !HAVE_READDIR_R 1486 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1487}
966 1488
967static void eio_atfork_child (void) 1489#define REQ(rtype) \
968{ 1490 eio_req *req; \
969 eio_req *prv; 1491 \
1492 req = (eio_req *)calloc (1, sizeof *req); \
1493 if (!req) \
1494 return 0; \
1495 \
1496 req->type = rtype; \
1497 req->pri = pri; \
1498 req->finish = cb; \
1499 req->data = data; \
1500 req->destroy = eio_api_destroy;
970 1501
971 while (prv = reqq_shift (&req_queue)) 1502#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1503
974 while (prv = reqq_shift (&res_queue)) 1504#define PATH \
975 eio_destroy (prv); 1505 req->flags |= EIO_FLAG_PTR1_FREE; \
1506 req->ptr1 = strdup (path); \
1507 if (!req->ptr1) \
1508 { \
1509 eio_api_destroy (req); \
1510 return 0; \
1511 }
976 1512
977 while (wrk_first.next != &wrk_first) 1513static void eio_execute (etp_worker *self, eio_req *req)
1514{
1515 errno = 0;
1516
1517 switch (req->type)
978 { 1518 {
979 worker *wrk = wrk_first.next; 1519 case EIO_READ: ALLOC (req->size);
1520 req->result = req->offs >= 0
1521 ? pread (req->int1, req->ptr2, req->size, req->offs)
1522 : read (req->int1, req->ptr2, req->size); break;
1523 case EIO_WRITE: req->result = req->offs >= 0
1524 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1525 : write (req->int1, req->ptr2, req->size); break;
980 1526
981 if (wrk->req) 1527 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1528 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1529
984 worker_clear (wrk); 1530 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1531 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1532 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1533 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1534 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1535 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1536
988 started = 0; 1537 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
989 idle = 0; 1538 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
990 nreqs = 0; 1539 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
991 nready = 0; 1540 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
992 npending = 0; 1541 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1542 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
993 1543
994 eio_atfork_parent (); 1544 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
995} 1545 case EIO_CLOSE: req->result = close (req->int1); break;
1546 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1547 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1548 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1549 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1550 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1551 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1552 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1553 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
996 1554
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1555 case EIO_READLINK: ALLOC (NAME_MAX);
998{ 1556 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
999 want_poll_cb = want_poll;
1000 done_poll_cb = done_poll;
1001 1557
1558 case EIO_SYNC: req->result = 0; sync (); break;
1559 case EIO_FSYNC: req->result = fsync (req->int1); break;
1560 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1561 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1562 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1563 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1564
1565 case EIO_READDIR: eio__scandir (req, self); break;
1566
1567 case EIO_BUSY:
1002#ifdef _WIN32 1568#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1569 Sleep (req->nv1 * 1000.);
1004 X_MUTEX_CHECK (reslock); 1570#else
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 { 1571 {
1130 dREQ; 1572 struct timeval tv;
1131 1573
1132 req->type = ix; 1574 tv.tv_sec = req->nv1;
1133 req->sv1 = newSVsv (fh); 1575 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1576
1142 if (!SvREADONLY (data)) 1577 req->result = select (0, 0, 0, 0, &tv);
1578 }
1579#endif
1580 break;
1581
1582 case EIO_UTIME:
1583 case EIO_FUTIME:
1584 {
1585 struct timeval tv[2];
1586 struct timeval *times;
1587
1588 if (req->nv1 != -1. || req->nv2 != -1.)
1143 { 1589 {
1144 SvREADONLY_on (data); 1590 tv[0].tv_sec = req->nv1;
1145 req->flags |= FLAG_SV2_RO_OFF; 1591 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1592 tv[1].tv_sec = req->nv2;
1593 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1594
1595 times = tv;
1146 } 1596 }
1597 else
1598 times = 0;
1147 1599
1148 EIO_SEND; 1600
1601 req->result = req->type == EIO_FUTIME
1602 ? futimes (req->int1, times)
1603 : utimes (req->ptr1, times);
1149 } 1604 }
1150} 1605 break;
1151 1606
1152void 1607 case EIO_GROUP:
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1608 abort (); /* handled in eio_request */
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159 1609
1160 data = newSV (NAME_MAX); 1610 case EIO_NOP:
1161 SvPOK_on (data); 1611 req->result = 0;
1612 break;
1162 1613
1163 req->type = EIO_READLINK; 1614 case EIO_CUSTOM:
1164 req->sv1 = newSVsv (path); 1615 ((void (*)(eio_req *))req->feed) (req);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1616 break;
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168 1617
1169 EIO_SEND; 1618 default:
1170} 1619 req->result = -1;
1620 break;
1621 }
1171 1622
1172void 1623 req->errorno = errno;
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188} 1624}
1189 1625
1190void 1626#ifndef EIO_NO_WRAPPERS
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196 1627
1197 req->type = EIO_READAHEAD; 1628eio_req *eio_nop (int pri, eio_cb cb, void *data)
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{ 1629{
1213 dREQ; 1630 REQ (EIO_NOP); SEND;
1631}
1214 1632
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT)); 1633eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216 if (!req->ptr2) 1634{
1217 { 1635 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218 req_destroy (req); 1636}
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221 1637
1638eio_req *eio_sync (int pri, eio_cb cb, void *data)
1639{
1640 REQ (EIO_SYNC); SEND;
1641}
1642
1643eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1644{
1645 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1646}
1647
1648eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1649{
1650 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1651}
1652
1653eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1654{
1655 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1656}
1657
1658eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1659{
1660 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1661}
1662
1663eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1664{
1665 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1666}
1667
1668eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1669{
1670 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1671}
1672
1673eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1674{
1675 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1676}
1677
1678eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1679{
1680 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1681}
1682
1683eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1684{
1685 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1686}
1687
1688eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1689{
1690 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1691}
1692
1693eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1694{
1695 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1696}
1697
1698eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1699{
1700 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1701}
1702
1703eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1704{
1705 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1706}
1707
1708eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1709{
1710 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1711}
1712
1713eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1714{
1715 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1716}
1717
1718eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1719{
1720 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1721}
1722
1723eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1724{
1725 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1726}
1727
1728eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1729{
1730 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1731}
1732
1733eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1734{
1735 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1736}
1737
1738eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1739{
1740 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1741}
1742
1743eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1744{
1745 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1746}
1747
1748eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1749{
1750 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1751}
1752
1753static eio_req *
1754eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1755{
1756 REQ (type); PATH; SEND;
1757}
1758
1759eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1760{
1761 return eio__1path (EIO_READLINK, path, pri, cb, data);
1762}
1763
1764eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1765{
1766 return eio__1path (EIO_STAT, path, pri, cb, data);
1767}
1768
1769eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1770{
1771 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1772}
1773
1774eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1775{
1776 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1777}
1778
1779eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1780{
1781 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1782}
1783
1784eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1785{
1786 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1787}
1788
1789eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1790{
1791 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1792}
1793
1794static eio_req *
1795eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1796{
1797 REQ (type); PATH;
1798
1222 req->flags |= FLAG_PTR2_FREE; 1799 req->flags |= EIO_FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path); 1800 req->ptr2 = strdup (new_path);
1801 if (!req->ptr2)
1802 {
1803 eio_api_destroy (req);
1804 return 0;
1805 }
1224 1806
1225 if (SvPOK (fh_or_path)) 1807 SEND;
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237} 1808}
1238 1809
1239void 1810eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{ 1811{
1243 dREQ; 1812 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261} 1813}
1262 1814
1263void 1815eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{ 1816{
1267 dREQ; 1817 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284} 1818}
1285 1819
1286void 1820eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{ 1821{
1294 dREQ; 1822 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306} 1823}
1307 1824
1308void 1825eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{ 1826{
1312 dREQ; 1827 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330} 1828}
1331 1829
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif 1830#endif
1831
1832eio_req *eio_grp (eio_cb cb, void *data)
1833{
1834 const int pri = EIO_PRI_MAX;
1835
1836 REQ (EIO_GROUP); SEND;
1837}
1838
1839#undef REQ
1840#undef PATH
1841#undef SEND
1842
1843/*****************************************************************************/
1844/* grp functions */
1418 1845
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 1846void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 1847{
1421 grp->int2 = limit; 1848 grp->int2 = limit;
1422 grp->feed = feed; 1849 grp->feed = feed;
1433 1860
1434void eio_grp_add (eio_req *grp, eio_req *req) 1861void eio_grp_add (eio_req *grp, eio_req *req)
1435{ 1862{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1863 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437 1864
1865 grp->flags |= EIO_FLAG_GROUPADD;
1866
1438 ++grp->size; 1867 ++grp->size;
1439 req->grp = grp; 1868 req->grp = grp;
1440 1869
1441 req->grp_prev = 0; 1870 req->grp_prev = 0;
1442 req->grp_next = grp->grp_first; 1871 req->grp_next = grp->grp_first;
1445 grp->grp_first->grp_prev = req; 1874 grp->grp_first->grp_prev = req;
1446 1875
1447 grp->grp_first = req; 1876 grp->grp_first = req;
1448} 1877}
1449 1878
1879/*****************************************************************************/
1880/* misc garbage */
1450 1881
1882ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1883{
1884 etp_worker wrk;
1885 ssize_t ret;
1886
1887 wrk.dbuf = 0;
1888
1889 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1890
1891 if (wrk.dbuf)
1892 free (wrk.dbuf);
1893
1894 return ret;
1895}
1896

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines