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 1.95 by root, Tue Jul 19 04:56:43 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
18 81
19#ifndef EIO_FINISH 82#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 84#endif
22 85
26 89
27#ifndef EIO_FEED 90#ifndef EIO_FEED
28# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
29#endif 92#endif
30 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
31#ifdef _WIN32 105#ifdef _WIN32
32 106
33 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 162
35#else 163#else
36 164
37# include "config.h"
38# include <sys/time.h> 165 #include <sys/time.h>
39# include <sys/select.h> 166 #include <sys/select.h>
167 #include <sys/statvfs.h>
40# include <unistd.h> 168 #include <unistd.h>
41# include <utime.h> 169 #include <utime.h>
42# include <signal.h> 170 #include <signal.h>
171 #include <dirent.h>
43 172
173 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
174 #include <sys/mman.h>
175 #endif
176
177 #define D_NAME(entp) entp->d_name
178
179 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
180 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
181 #define _DIRENT_HAVE_D_TYPE /* sigh */
182 #define D_INO(de) (de)->d_fileno
183 #define D_NAMLEN(de) (de)->d_namlen
184 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
185 #define D_INO(de) (de)->d_ino
186 #endif
187
188 #ifdef _D_EXACT_NAMLEN
189 #undef D_NAMLEN
190 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
191 #endif
192
193 #ifdef _DIRENT_HAVE_D_TYPE
194 #define D_TYPE(de) (de)->d_type
195 #endif
196
44# ifndef EIO_STRUCT_DIRENT 197 #ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 198 #define EIO_STRUCT_DIRENT struct dirent
199 #endif
200
46# endif 201#endif
47
48#endif
49
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53 202
54#if HAVE_SENDFILE 203#if HAVE_SENDFILE
55# if __linux 204# if __linux
56# include <sys/sendfile.h> 205# include <sys/sendfile.h>
57# elif __freebsd 206# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 207# include <sys/socket.h>
59# include <sys/uio.h> 208# include <sys/uio.h>
60# elif __hpux 209# elif __hpux
61# include <sys/socket.h> 210# include <sys/socket.h>
62# elif __solaris /* not yet */ 211# elif __solaris
63# include <sys/sendfile.h> 212# include <sys/sendfile.h>
64# else 213# else
65# error sendfile support requested but not available 214# error sendfile support requested but not available
66# endif 215# endif
67#endif 216#endif
68 217
69/* number of seconds after which an idle threads exit */ 218#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 219# define D_TYPE(de) 0
220#endif
221#ifndef D_INO
222# define D_INO(de) 0
223#endif
224#ifndef D_NAMLEN
225# define D_NAMLEN(entp) strlen (D_NAME (entp))
226#endif
71 227
72/* used for struct dirent, AIX doesn't provide it */ 228/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 229#ifndef NAME_MAX
74# define NAME_MAX 4096 230# define NAME_MAX 4096
75#endif 231#endif
76 232
233/* used for readlink etc. */
234#ifndef PATH_MAX
235# define PATH_MAX 4096
236#endif
237
77/* buffer size for various temporary buffers */ 238/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 239#define EIO_BUFSIZE 65536
79 240
80#define dBUF \ 241#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 242 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 243 errno = ENOMEM; \
85 if (!eio_buf) \ 244 if (!eio_buf) \
86 return -1; 245 return -1
246
247#define FUBd \
248 free (eio_buf)
87 249
88#define EIO_TICKS ((1000000 + 1023) >> 10) 250#define EIO_TICKS ((1000000 + 1023) >> 10)
251
252#define ETP_PRI_MIN EIO_PRI_MIN
253#define ETP_PRI_MAX EIO_PRI_MAX
254
255struct etp_worker;
256
257#define ETP_REQ eio_req
258#define ETP_DESTROY(req) eio_destroy (req)
259static int eio_finish (eio_req *req);
260#define ETP_FINISH(req) eio_finish (req)
261static void eio_execute (struct etp_worker *self, eio_req *req);
262#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
263
264/*****************************************************************************/
265
266#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
267
268/* calculate time difference in ~1/EIO_TICKS of a second */
269ecb_inline int
270tvdiff (struct timeval *tv1, struct timeval *tv2)
271{
272 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
273 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
274}
275
276static unsigned int started, idle, wanted = 4;
89 277
90static void (*want_poll_cb) (void); 278static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 279static void (*done_poll_cb) (void);
92 280
93static unsigned int max_poll_time = 0; 281static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 282static unsigned int max_poll_reqs; /* reslock */
95 283
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 284static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 285static volatile unsigned int nready; /* reqlock */
98{ 286static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 287static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 288static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 289
103static unsigned int started, idle, wanted; 290static xmutex_t wrklock;
291static xmutex_t reslock;
292static xmutex_t reqlock;
293static xcond_t reqwait;
294
295#if !HAVE_PREADWRITE
296/*
297 * make our pread/pwrite emulation safe against themselves, but not against
298 * normal read/write by using a mutex. slows down execution a lot,
299 * but that's your problem, not mine.
300 */
301static xmutex_t preadwritelock;
302#endif
303
304typedef struct etp_worker
305{
306 /* locked by wrklock */
307 struct etp_worker *prev, *next;
308
309 xthread_t tid;
310
311 /* locked by reslock, reqlock or wrklock */
312 ETP_REQ *req; /* currently processed request */
313
314#ifdef ETP_WORKER_COMMON
315 ETP_WORKER_COMMON
316#endif
317} etp_worker;
318
319static etp_worker wrk_first; /* NOT etp */
320
321#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
322#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 323
105/* worker threads management */ 324/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 325
108typedef struct worker { 326static void ecb_cold
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 327etp_worker_clear (etp_worker *wrk)
123{ 328{
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} 329}
136 330
137static void worker_free (worker *wrk) 331static void ecb_cold
332etp_worker_free (etp_worker *wrk)
138{ 333{
139 wrk->next->prev = wrk->prev; 334 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 335 wrk->prev->next = wrk->next;
141 336
142 free (wrk); 337 free (wrk);
143} 338}
144 339
145static volatile unsigned int nreqs, nready, npending; 340static unsigned int
146static volatile unsigned int max_idle = 4; 341etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 342{
343 int retval;
344 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
345 retval = nreqs;
346 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 347 return retval;
155} 348}
156 349
157unsigned int eio_nready (void) 350static unsigned int
351etp_nready (void)
158{ 352{
159 unsigned int retval; 353 unsigned int retval;
160 354
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 356 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 358
165 return retval; 359 return retval;
166} 360}
167 361
168unsigned int eio_npending (void) 362static unsigned int
363etp_npending (void)
169{ 364{
170 unsigned int retval; 365 unsigned int retval;
171 366
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 367 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 368 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 370
176 return retval; 371 return retval;
177} 372}
178 373
179unsigned int eio_nthreads (void) 374static unsigned int
375etp_nthreads (void)
180{ 376{
181 unsigned int retval; 377 unsigned int retval;
182 378
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 379 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 380 retval = started;
191 * a somewhat faster data structure might be nice, but 387 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 388 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 389 * per shift, the most expensive operation.
194 */ 390 */
195typedef struct { 391typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 392 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 393 int size;
198} reqq; 394} etp_reqq;
199 395
200static reqq req_queue; 396static etp_reqq req_queue;
201static reqq res_queue; 397static etp_reqq res_queue;
202 398
203static int reqq_push (reqq *q, eio_req *req) 399static void ecb_noinline ecb_cold
400reqq_init (etp_reqq *q)
401{
402 int pri;
403
404 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
405 q->qs[pri] = q->qe[pri] = 0;
406
407 q->size = 0;
408}
409
410static int ecb_noinline
411reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 412{
205 int pri = req->pri; 413 int pri = req->pri;
206 req->next = 0; 414 req->next = 0;
207 415
208 if (q->qe[pri]) 416 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 422 q->qe[pri] = q->qs[pri] = req;
215 423
216 return q->size++; 424 return q->size++;
217} 425}
218 426
219static eio_req *reqq_shift (reqq *q) 427static ETP_REQ * ecb_noinline
428reqq_shift (etp_reqq *q)
220{ 429{
221 int pri; 430 int pri;
222 431
223 if (!q->size) 432 if (!q->size)
224 return 0; 433 return 0;
225 434
226 --q->size; 435 --q->size;
227 436
228 for (pri = EIO_NUM_PRI; pri--; ) 437 for (pri = ETP_NUM_PRI; pri--; )
229 { 438 {
230 eio_req *req = q->qs[pri]; 439 eio_req *req = q->qs[pri];
231 440
232 if (req) 441 if (req)
233 { 442 {
239 } 448 }
240 449
241 abort (); 450 abort ();
242} 451}
243 452
244static void grp_try_feed (eio_req *grp) 453static int ecb_cold
454etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 455{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 456 X_MUTEX_CREATE (wrklock);
247 { 457 X_MUTEX_CREATE (reslock);
248 int old_len = grp->size; 458 X_MUTEX_CREATE (reqlock);
459 X_COND_CREATE (reqwait);
249 460
250 EIO_FEED (grp); 461 reqq_init (&req_queue);
462 reqq_init (&res_queue);
251 463
252 /* stop if no progress has been made */ 464 wrk_first.next =
253 if (old_len == grp->size) 465 wrk_first.prev = &wrk_first;
254 {
255 grp->feed = 0;
256 break;
257 }
258 }
259}
260 466
261static int eio_finish (eio_req *req); 467 started = 0;
468 idle = 0;
469 nreqs = 0;
470 nready = 0;
471 npending = 0;
262 472
263static int grp_dec (eio_req *grp) 473 want_poll_cb = want_poll;
264{ 474 done_poll_cb = done_poll;
265 --grp->size;
266 475
267 /* call feeder, if applicable */
268 grp_try_feed (grp);
269
270 /* finish, if done */
271 if (!grp->size && grp->int1)
272 return eio_finish (grp);
273 else
274 return 0; 476 return 0;
275} 477}
276 478
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); 479X_THREAD_PROC (etp_proc);
326 480
327static void start_thread (void) 481static void ecb_cold
482etp_start_thread (void)
328{ 483{
329 worker *wrk = calloc (1, sizeof (worker)); 484 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 485
331 if (!wrk) 486 /*TODO*/
332 croak ("unable to allocate worker thread data"); 487 assert (("unable to allocate worker thread data", wrk));
333 488
334 X_LOCK (wrklock); 489 X_LOCK (wrklock);
335 490
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 491 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 492 {
338 wrk->prev = &wrk_first; 493 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 494 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 495 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 496 wrk_first.next = wrk;
345 free (wrk); 500 free (wrk);
346 501
347 X_UNLOCK (wrklock); 502 X_UNLOCK (wrklock);
348} 503}
349 504
505static void
350static void maybe_start_thread (void) 506etp_maybe_start_thread (void)
351{ 507{
352 if (eio_nthreads () >= wanted) 508 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 509 return;
354 510
355 /* todo: maybe use idle here, but might be less exact */ 511 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 512 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 513 return;
358 514
359 start_thread (); 515 etp_start_thread ();
360} 516}
361 517
362void eio_submit (eio_req *req) 518static void ecb_cold
363{ 519etp_end_thread (void)
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{ 520{
377 eio_req *req = calloc (1, sizeof (eio_req)); 521 eio_req *req = calloc (1, sizeof (eio_req));
378 522
379 req->type = EIO_QUIT; 523 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 524 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 525
382 X_LOCK (reqlock); 526 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 527 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 528 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 529 X_UNLOCK (reqlock);
387 X_LOCK (wrklock); 531 X_LOCK (wrklock);
388 --started; 532 --started;
389 X_UNLOCK (wrklock); 533 X_UNLOCK (wrklock);
390} 534}
391 535
392void eio_set_max_poll_time (double nseconds) 536static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 537etp_poll (void)
429{ 538{
430 int maxreqs = max_poll_reqs; 539 unsigned int maxreqs;
540 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 541 struct timeval tv_start, tv_now;
432 eio_req *req;
433 542
543 X_LOCK (reslock);
544 maxreqs = max_poll_reqs;
545 maxtime = max_poll_time;
546 X_UNLOCK (reslock);
547
434 if (max_poll_time) 548 if (maxtime)
435 gettimeofday (&tv_start, 0); 549 gettimeofday (&tv_start, 0);
436 550
437 for (;;) 551 for (;;)
438 { 552 {
553 ETP_REQ *req;
554
439 maybe_start_thread (); 555 etp_maybe_start_thread ();
440 556
441 X_LOCK (reslock); 557 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 558 req = reqq_shift (&res_queue);
443 559
444 if (req) 560 if (req)
452 X_UNLOCK (reslock); 568 X_UNLOCK (reslock);
453 569
454 if (!req) 570 if (!req)
455 return 0; 571 return 0;
456 572
573 X_LOCK (reqlock);
457 --nreqs; 574 --nreqs;
575 X_UNLOCK (reqlock);
458 576
459 if (req->type == EIO_GROUP && req->size) 577 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 578 {
461 req->int1 = 1; /* mark request as delayed */ 579 req->int1 = 1; /* mark request as delayed */
462 continue; 580 continue;
463 } 581 }
464 else 582 else
465 { 583 {
466 int res = eio_finish (req); 584 int res = ETP_FINISH (req);
467 if (res) 585 if (ecb_expect_false (res))
468 return res; 586 return res;
469 } 587 }
470 588
471 if (maxreqs && !--maxreqs) 589 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 590 break;
473 591
474 if (max_poll_time) 592 if (maxtime)
475 { 593 {
476 gettimeofday (&tv_now, 0); 594 gettimeofday (&tv_now, 0);
477 595
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 596 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 597 break;
480 } 598 }
481 } 599 }
482 600
483 errno = EAGAIN; 601 errno = EAGAIN;
484 return -1; 602 return -1;
485} 603}
486 604
605static void
606etp_cancel (ETP_REQ *req)
607{
608 req->cancelled = 1;
609
610 eio_grp_cancel (req);
611}
612
613static void
614etp_submit (ETP_REQ *req)
615{
616 req->pri -= ETP_PRI_MIN;
617
618 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
619 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
620
621 if (ecb_expect_false (req->type == EIO_GROUP))
622 {
623 /* I hope this is worth it :/ */
624 X_LOCK (reqlock);
625 ++nreqs;
626 X_UNLOCK (reqlock);
627
628 X_LOCK (reslock);
629
630 ++npending;
631
632 if (!reqq_push (&res_queue, req) && want_poll_cb)
633 want_poll_cb ();
634
635 X_UNLOCK (reslock);
636 }
637 else
638 {
639 X_LOCK (reqlock);
640 ++nreqs;
641 ++nready;
642 reqq_push (&req_queue, req);
643 X_COND_SIGNAL (reqwait);
644 X_UNLOCK (reqlock);
645
646 etp_maybe_start_thread ();
647 }
648}
649
650static void ecb_cold
651etp_set_max_poll_time (double nseconds)
652{
653 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
654 max_poll_time = nseconds * EIO_TICKS;
655 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
656}
657
658static void ecb_cold
659etp_set_max_poll_reqs (unsigned int maxreqs)
660{
661 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
662 max_poll_reqs = maxreqs;
663 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
664}
665
666static void ecb_cold
667etp_set_max_idle (unsigned int nthreads)
668{
669 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
670 max_idle = nthreads;
671 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
672}
673
674static void ecb_cold
675etp_set_idle_timeout (unsigned int seconds)
676{
677 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
678 idle_timeout = seconds;
679 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
680}
681
682static void ecb_cold
683etp_set_min_parallel (unsigned int nthreads)
684{
685 if (wanted < nthreads)
686 wanted = nthreads;
687}
688
689static void ecb_cold
690etp_set_max_parallel (unsigned int nthreads)
691{
692 if (wanted > nthreads)
693 wanted = nthreads;
694
695 while (started > wanted)
696 etp_end_thread ();
697}
698
699/*****************************************************************************/
700
701static void
702grp_try_feed (eio_req *grp)
703{
704 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
705 {
706 grp->flags &= ~EIO_FLAG_GROUPADD;
707
708 EIO_FEED (grp);
709
710 /* stop if no progress has been made */
711 if (!(grp->flags & EIO_FLAG_GROUPADD))
712 {
713 grp->feed = 0;
714 break;
715 }
716 }
717}
718
719static int
720grp_dec (eio_req *grp)
721{
722 --grp->size;
723
724 /* call feeder, if applicable */
725 grp_try_feed (grp);
726
727 /* finish, if done */
728 if (!grp->size && grp->int1)
729 return eio_finish (grp);
730 else
731 return 0;
732}
733
734static void
735eio_destroy (eio_req *req)
736{
737 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
738 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
739
740 EIO_DESTROY (req);
741}
742
743static int
744eio_finish (eio_req *req)
745{
746 int res = EIO_FINISH (req);
747
748 if (req->grp)
749 {
750 int res2;
751 eio_req *grp = req->grp;
752
753 /* unlink request */
754 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
755 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
756
757 if (grp->grp_first == req)
758 grp->grp_first = req->grp_next;
759
760 res2 = grp_dec (grp);
761
762 if (!res)
763 res = res2;
764 }
765
766 eio_destroy (req);
767
768 return res;
769}
770
771void
772eio_grp_cancel (eio_req *grp)
773{
774 for (grp = grp->grp_first; grp; grp = grp->grp_next)
775 eio_cancel (grp);
776}
777
778void
779eio_cancel (eio_req *req)
780{
781 etp_cancel (req);
782}
783
784void
785eio_submit (eio_req *req)
786{
787 etp_submit (req);
788}
789
790unsigned int
791eio_nreqs (void)
792{
793 return etp_nreqs ();
794}
795
796unsigned int
797eio_nready (void)
798{
799 return etp_nready ();
800}
801
802unsigned int
803eio_npending (void)
804{
805 return etp_npending ();
806}
807
808unsigned int ecb_cold
809eio_nthreads (void)
810{
811 return etp_nthreads ();
812}
813
814void ecb_cold
815eio_set_max_poll_time (double nseconds)
816{
817 etp_set_max_poll_time (nseconds);
818}
819
820void ecb_cold
821eio_set_max_poll_reqs (unsigned int maxreqs)
822{
823 etp_set_max_poll_reqs (maxreqs);
824}
825
826void ecb_cold
827eio_set_max_idle (unsigned int nthreads)
828{
829 etp_set_max_idle (nthreads);
830}
831
832void ecb_cold
833eio_set_idle_timeout (unsigned int seconds)
834{
835 etp_set_idle_timeout (seconds);
836}
837
838void ecb_cold
839eio_set_min_parallel (unsigned int nthreads)
840{
841 etp_set_min_parallel (nthreads);
842}
843
844void ecb_cold
845eio_set_max_parallel (unsigned int nthreads)
846{
847 etp_set_max_parallel (nthreads);
848}
849
850int eio_poll (void)
851{
852 return etp_poll ();
853}
854
487/*****************************************************************************/ 855/*****************************************************************************/
488/* work around various missing functions */ 856/* work around various missing functions */
489 857
490#if !HAVE_PREADWRITE 858#if !HAVE_PREADWRITE
859# undef pread
860# undef pwrite
491# define pread aio_pread 861# define pread eio__pread
492# define pwrite aio_pwrite 862# define pwrite eio__pwrite
493 863
494/* 864static eio_ssize_t
495 * make our pread/pwrite safe against themselves, but not against 865eio__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{ 866{
503 ssize_t res; 867 eio_ssize_t res;
504 off_t ooffset; 868 off_t ooffset;
505 869
506 X_LOCK (preadwritelock); 870 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR); 871 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET); 872 lseek (fd, offset, SEEK_SET);
511 X_UNLOCK (preadwritelock); 875 X_UNLOCK (preadwritelock);
512 876
513 return res; 877 return res;
514} 878}
515 879
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 880static eio_ssize_t
881eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 882{
518 ssize_t res; 883 eio_ssize_t res;
519 off_t ooffset; 884 off_t ooffset;
520 885
521 X_LOCK (preadwritelock); 886 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 887 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 888 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 889 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 890 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 891 X_UNLOCK (preadwritelock);
527 892
528 return res; 893 return res;
529} 894}
530#endif 895#endif
531 896
532#ifndef HAVE_FUTIMES 897#ifndef HAVE_UTIMES
533 898
899# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 900# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 901
902static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 903eio__utimes (const char *filename, const struct timeval times[2])
538{ 904{
539 if (times) 905 if (times)
540 { 906 {
541 struct utimbuf buf; 907 struct utimbuf buf;
542 908
547 } 913 }
548 else 914 else
549 return utime (filename, 0); 915 return utime (filename, 0);
550} 916}
551 917
918#endif
919
920#ifndef HAVE_FUTIMES
921
922# undef futimes
923# define futimes(fd,times) eio__futimes (fd, times)
924
925static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 926eio__futimes (int fd, const struct timeval tv[2])
553{ 927{
554 errno = ENOSYS; 928 errno = ENOSYS;
555 return -1; 929 return -1;
556} 930}
557 931
558#endif 932#endif
559 933
560#if !HAVE_FDATASYNC 934#if !HAVE_FDATASYNC
935# undef fdatasync
561# define fdatasync fsync 936# define fdatasync(fd) fsync (fd)
562#endif 937#endif
938
939/* sync_file_range always needs emulation */
940static int
941eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
942{
943#if HAVE_SYNC_FILE_RANGE
944 int res;
945
946 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
947 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
948 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
949 {
950 flags = 0
951 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
952 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
953 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
954 }
955
956 res = sync_file_range (fd, offset, nbytes, flags);
957
958 if (!res || errno != ENOSYS)
959 return res;
960#endif
961
962 /* even though we could play tricks with the flags, it's better to always
963 * call fdatasync, as that matches the expectation of its users best */
964 return fdatasync (fd);
965}
966
967static int
968eio__fallocate (int fd, int mode, off_t offset, size_t len)
969{
970#if HAVE_FALLOCATE
971 return fallocate (fd, mode, offset, len);
972#else
973 errno = ENOSYS;
974 return -1;
975#endif
976}
563 977
564#if !HAVE_READAHEAD 978#if !HAVE_READAHEAD
979# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 980# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 981
982static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 983eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 984{
569 size_t todo = count; 985 size_t todo = count;
570 dBUF; 986 dBUF;
571 987
572 while (todo > 0) 988 while (todo > 0)
573 { 989 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 990 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 991
576 pread (fd, aio_buf, len, offset); 992 pread (fd, eio_buf, len, offset);
577 offset += len; 993 offset += len;
578 todo -= len; 994 todo -= len;
579 } 995 }
580 996
997 FUBd;
998
581 errno = 0; 999 errno = 0;
582 return count; 1000 return count;
583} 1001}
584 1002
585#endif 1003#endif
586 1004
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 */ 1005/* sendfile always needs emulation */
1006static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1007eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 1008{
1009 eio_ssize_t written = 0;
620 ssize_t res; 1010 eio_ssize_t res;
621 1011
622 if (!count) 1012 if (!count)
623 return 0; 1013 return 0;
624 1014
1015 for (;;)
1016 {
1017#ifdef __APPLE__
1018# undef HAVE_SENDFILE /* broken, as everything on os x */
1019#endif
625#if HAVE_SENDFILE 1020#if HAVE_SENDFILE
626# if __linux 1021# if __linux
1022 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1023 res = sendfile (ofd, ifd, &soffset, count);
628 1024
629# elif __freebsd 1025# elif __FreeBSD__
630 /* 1026 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1027 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1028 * wasted on making it similar to other I/O functions.
633 */ 1029 */
634 {
635 off_t sbytes; 1030 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1031 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1032
638 if (res < 0 && sbytes) 1033 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1034 /* freebsd' sendfile will return 0 on success */
1035 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1036 /* not on e.g. EIO or EPIPE - sounds broken */
1037 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1038 res = sbytes;
641 } 1039 #endif
1040
1041 /* according to source inspection, this is correct, and useful behaviour */
1042 if (sbytes)
1043 res = sbytes;
1044
1045# elif defined (__APPLE__)
1046 off_t sbytes = count;
1047 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1048
1049 /* according to the manpage, sbytes is always valid */
1050 if (sbytes)
1051 res = sbytes;
642 1052
643# elif __hpux 1053# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1054 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1055
646# elif __solaris 1056# elif __solaris
647 {
648 struct sendfilevec vec; 1057 struct sendfilevec vec;
649 size_t sbytes; 1058 size_t sbytes;
650 1059
651 vec.sfv_fd = ifd; 1060 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1061 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1062 vec.sfv_off = offset;
654 vec.sfv_len = count; 1063 vec.sfv_len = count;
655 1064
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1065 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1066
658 if (res < 0 && sbytes) 1067 if (res < 0 && sbytes)
659 res = sbytes; 1068 res = sbytes;
660 }
661 1069
662# endif 1070# endif
1071
1072#elif defined (_WIN32) && 0
1073 /* does not work, just for documentation of what would need to be done */
1074 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1075 /* libeio guarantees that the file offset does not change, and windows */
1076 /* has no way to get an independent handle to the same file description */
1077 HANDLE h = TO_SOCKET (ifd);
1078 SetFilePointer (h, offset, 0, FILE_BEGIN);
1079 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1080
663#else 1081#else
664 res = -1; 1082 res = -1;
665 errno = ENOSYS; 1083 errno = ENOSYS;
666#endif 1084#endif
667 1085
1086 /* we assume sendfile can copy at least 128mb in one go */
1087 if (res <= 128 * 1024 * 1024)
1088 {
1089 if (res > 0)
1090 written += res;
1091
1092 if (written)
1093 return written;
1094
1095 break;
1096 }
1097 else
1098 {
1099 /* if we requested more, then probably the kernel was lazy */
1100 written += res;
1101 offset += res;
1102 count -= res;
1103
1104 if (!count)
1105 return written;
1106 }
1107 }
1108
668 if (res < 0 1109 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1110 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1111 /* BSDs */
1112#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1113 || errno == ENOTSUP
1114#endif
1115#ifdef EOPNOTSUPP /* windows */
1116 || errno == EOPNOTSUPP /* BSDs */
1117#endif
670#if __solaris 1118#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1119 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1120#endif
673 ) 1121 )
674 ) 1122 )
678 1126
679 res = 0; 1127 res = 0;
680 1128
681 while (count) 1129 while (count)
682 { 1130 {
683 ssize_t cnt; 1131 eio_ssize_t cnt;
684 1132
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1133 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 1134
687 if (cnt <= 0) 1135 if (cnt <= 0)
688 { 1136 {
700 1148
701 offset += cnt; 1149 offset += cnt;
702 res += cnt; 1150 res += cnt;
703 count -= cnt; 1151 count -= cnt;
704 } 1152 }
1153
1154 FUBd;
705 } 1155 }
706 1156
707 return res; 1157 return res;
708} 1158}
709 1159
1160#ifdef PAGESIZE
1161# define eio_pagesize() PAGESIZE
1162#else
1163static intptr_t
1164eio_pagesize (void)
1165{
1166 static intptr_t page;
1167
1168 if (!page)
1169 page = sysconf (_SC_PAGESIZE);
1170
1171 return page;
1172}
1173#endif
1174
1175static void
1176eio_page_align (void **addr, size_t *length)
1177{
1178 intptr_t mask = eio_pagesize () - 1;
1179
1180 /* round down addr */
1181 intptr_t adj = mask & (intptr_t)*addr;
1182
1183 *addr = (void *)((intptr_t)*addr - adj);
1184 *length += adj;
1185
1186 /* round up length */
1187 *length = (*length + mask) & ~mask;
1188}
1189
1190#if !_POSIX_MEMLOCK
1191# define eio__mlockall(a) EIO_ENOSYS ()
1192#else
1193
1194static int
1195eio__mlockall (int flags)
1196{
1197 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1198 extern int mallopt (int, int);
1199 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1200 #endif
1201
1202 if (EIO_MCL_CURRENT != MCL_CURRENT
1203 || EIO_MCL_FUTURE != MCL_FUTURE)
1204 {
1205 flags = 0
1206 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1207 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1208 }
1209
1210 return mlockall (flags);
1211}
1212#endif
1213
1214#if !_POSIX_MEMLOCK_RANGE
1215# define eio__mlock(a,b) EIO_ENOSYS ()
1216#else
1217
1218static int
1219eio__mlock (void *addr, size_t length)
1220{
1221 eio_page_align (&addr, &length);
1222
1223 return mlock (addr, length);
1224}
1225
1226#endif
1227
1228#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1229# define eio__msync(a,b,c) EIO_ENOSYS ()
1230#else
1231
1232static int
1233eio__msync (void *mem, size_t len, int flags)
1234{
1235 eio_page_align (&mem, &len);
1236
1237 if (EIO_MS_ASYNC != MS_SYNC
1238 || EIO_MS_INVALIDATE != MS_INVALIDATE
1239 || EIO_MS_SYNC != MS_SYNC)
1240 {
1241 flags = 0
1242 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1243 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1244 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1245 }
1246
1247 return msync (mem, len, flags);
1248}
1249
1250#endif
1251
1252static int
1253eio__mtouch (eio_req *req)
1254{
1255 void *mem = req->ptr2;
1256 size_t len = req->size;
1257 int flags = req->int1;
1258
1259 eio_page_align (&mem, &len);
1260
1261 {
1262 intptr_t addr = (intptr_t)mem;
1263 intptr_t end = addr + len;
1264 intptr_t page = eio_pagesize ();
1265
1266 if (addr < end)
1267 if (flags & EIO_MT_MODIFY) /* modify */
1268 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1269 else
1270 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1271 }
1272
1273 return 0;
1274}
1275
1276/*****************************************************************************/
1277/* requests implemented outside eio_execute, because they are so large */
1278
1279static void
1280eio__realpath (eio_req *req, etp_worker *self)
1281{
1282 char *rel = req->ptr1;
1283 char *res;
1284 char *tmp1, *tmp2;
1285#if SYMLOOP_MAX > 32
1286 int symlinks = SYMLOOP_MAX;
1287#else
1288 int symlinks = 32;
1289#endif
1290
1291 req->result = -1;
1292
1293 errno = EINVAL;
1294 if (!rel)
1295 return;
1296
1297 errno = ENOENT;
1298 if (!*rel)
1299 return;
1300
1301 if (!req->ptr2)
1302 {
1303 X_LOCK (wrklock);
1304 req->flags |= EIO_FLAG_PTR2_FREE;
1305 X_UNLOCK (wrklock);
1306 req->ptr2 = malloc (PATH_MAX * 3);
1307
1308 errno = ENOMEM;
1309 if (!req->ptr2)
1310 return;
1311 }
1312
1313 res = req->ptr2;
1314 tmp1 = res + PATH_MAX;
1315 tmp2 = tmp1 + PATH_MAX;
1316
1317#if 0 /* disabled, the musl way to do things is just too racy */
1318#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1319 /* on linux we may be able to ask the kernel */
1320 {
1321 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1322
1323 if (fd >= 0)
1324 {
1325 sprintf (tmp1, "/proc/self/fd/%d", fd);
1326 req->result = readlink (tmp1, res, PATH_MAX);
1327 close (fd);
1328
1329 /* here we should probably stat the open file and the disk file, to make sure they still match */
1330
1331 if (req->result > 0)
1332 goto done;
1333 }
1334 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1335 return;
1336 }
1337#endif
1338#endif
1339
1340 if (*rel != '/')
1341 {
1342 if (!getcwd (res, PATH_MAX))
1343 return;
1344
1345 if (res [1]) /* only use if not / */
1346 res += strlen (res);
1347 }
1348
1349 while (*rel)
1350 {
1351 eio_ssize_t len, linklen;
1352 char *beg = rel;
1353
1354 while (*rel && *rel != '/')
1355 ++rel;
1356
1357 len = rel - beg;
1358
1359 if (!len) /* skip slashes */
1360 {
1361 ++rel;
1362 continue;
1363 }
1364
1365 if (beg [0] == '.')
1366 {
1367 if (len == 1)
1368 continue; /* . - nop */
1369
1370 if (beg [1] == '.' && len == 2)
1371 {
1372 /* .. - back up one component, if possible */
1373
1374 while (res != req->ptr2)
1375 if (*--res == '/')
1376 break;
1377
1378 continue;
1379 }
1380 }
1381
1382 errno = ENAMETOOLONG;
1383 if (res + 1 + len + 1 >= tmp1)
1384 return;
1385
1386 /* copy one component */
1387 *res = '/';
1388 memcpy (res + 1, beg, len);
1389
1390 /* zero-terminate, for readlink */
1391 res [len + 1] = 0;
1392
1393 /* now check if it's a symlink */
1394 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1395
1396 if (linklen < 0)
1397 {
1398 if (errno != EINVAL)
1399 return;
1400
1401 /* it's a normal directory. hopefully */
1402 res += len + 1;
1403 }
1404 else
1405 {
1406 /* yay, it was a symlink - build new path in tmp2 */
1407 int rellen = strlen (rel);
1408
1409 errno = ENAMETOOLONG;
1410 if (linklen + 1 + rellen >= PATH_MAX)
1411 return;
1412
1413 errno = ELOOP;
1414 if (!--symlinks)
1415 return;
1416
1417 if (*tmp1 == '/')
1418 res = req->ptr2; /* symlink resolves to an absolute path */
1419
1420 /* we need to be careful, as rel might point into tmp2 already */
1421 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1422 tmp2 [linklen] = '/';
1423 memcpy (tmp2, tmp1, linklen);
1424
1425 rel = tmp2;
1426 }
1427 }
1428
1429 /* special case for the lone root path */
1430 if (res == req->ptr2)
1431 *res++ = '/';
1432
1433 req->result = res - (char *)req->ptr2;
1434
1435done:
1436 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1437}
1438
1439static signed char
1440eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1441{
1442 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1443 : a->inode < b->inode ? -1
1444 : a->inode > b->inode ? 1
1445 : 0;
1446}
1447
1448#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1449
1450#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1451#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1452
1453static void
1454eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1455{
1456 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1457 unsigned char *bit = bits;
1458
1459 assert (CHAR_BIT == 8);
1460 assert (sizeof (eio_dirent) * 8 < 256);
1461 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1462 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1463
1464 if (size <= EIO_SORT_FAST)
1465 return;
1466
1467 /* first prepare an array of bits to test in our radix sort */
1468 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1469 /* inode_bits must contain all inodes ORed together */
1470 /* which is used to skip bits that are 0 everywhere, which is very common */
1471 {
1472 eio_ino_t endianness;
1473 int i, j;
1474
1475 /* we store the byte offset of byte n into byte n of "endianness" */
1476 for (i = 0; i < sizeof (eio_ino_t); ++i)
1477 ((unsigned char *)&endianness)[i] = i;
1478
1479 *bit++ = 0;
1480
1481 for (i = 0; i < sizeof (eio_ino_t); ++i)
1482 {
1483 /* shifting off the byte offsets out of "endianness" */
1484 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1485 endianness >>= 8;
1486
1487 for (j = 0; j < 8; ++j)
1488 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1489 *bit++ = offs + j;
1490 }
1491
1492 for (j = 0; j < 8; ++j)
1493 if (score_bits & (1 << j))
1494 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1495 }
1496
1497 /* now actually do the sorting (a variant of MSD radix sort) */
1498 {
1499 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1500 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1501 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1502 int stk_idx = 0;
1503
1504 base_stk [stk_idx] = dents;
1505 end_stk [stk_idx] = dents + size;
1506 bit_stk [stk_idx] = bit - 1;
1507
1508 do
1509 {
1510 base = base_stk [stk_idx];
1511 end = end_stk [stk_idx];
1512 bit = bit_stk [stk_idx];
1513
1514 for (;;)
1515 {
1516 unsigned char O = *bit >> 3;
1517 unsigned char M = 1 << (*bit & 7);
1518
1519 eio_dirent *a = base;
1520 eio_dirent *b = end;
1521
1522 if (b - a < EIO_SORT_CUTOFF)
1523 break;
1524
1525 /* now bit-partition the array on the bit */
1526 /* this ugly asymmetric loop seems to perform much better than typical */
1527 /* partition algos found in the literature */
1528 do
1529 if (!(((unsigned char *)a)[O] & M))
1530 ++a;
1531 else if (!(((unsigned char *)--b)[O] & M))
1532 {
1533 eio_dirent tmp = *a; *a = *b; *b = tmp;
1534 ++a;
1535 }
1536 while (b > a);
1537
1538 /* next bit, or stop, if no bits left in this path */
1539 if (!*--bit)
1540 break;
1541
1542 base_stk [stk_idx] = a;
1543 end_stk [stk_idx] = end;
1544 bit_stk [stk_idx] = bit;
1545 ++stk_idx;
1546
1547 end = a;
1548 }
1549 }
1550 while (stk_idx--);
1551 }
1552}
1553
1554static void
1555eio_dent_insertion_sort (eio_dirent *dents, int size)
1556{
1557 /* first move the smallest element to the front, to act as a sentinel */
1558 {
1559 int i;
1560 eio_dirent *min = dents;
1561
1562 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1563 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1564 if (EIO_DENT_CMP (dents [i], <, *min))
1565 min = &dents [i];
1566
1567 /* swap elements 0 and j (minimum) */
1568 {
1569 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1570 }
1571 }
1572
1573 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1574 {
1575 eio_dirent *i, *j;
1576
1577 for (i = dents + 1; i < dents + size; ++i)
1578 {
1579 eio_dirent value = *i;
1580
1581 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1582 j [1] = j [0];
1583
1584 j [1] = value;
1585 }
1586 }
1587}
1588
1589static void
1590eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1591{
1592 if (size <= 1)
1593 return; /* our insertion sort relies on size > 0 */
1594
1595 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1596 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1597 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1598
1599 /* use an insertion sort at the end, or for small arrays, */
1600 /* as insertion sort is more efficient for small partitions */
1601 eio_dent_insertion_sort (dents, size);
1602}
1603
710/* read a full directory */ 1604/* read a full directory */
1605static void
711static void scandir_ (eio_req *req, worker *self) 1606eio__scandir (eio_req *req, etp_worker *self)
712{ 1607{
1608 char *name, *names;
1609 int namesalloc = 4096 - sizeof (void *) * 4;
1610 int namesoffs = 0;
1611 int flags = req->int1;
1612 eio_dirent *dents = 0;
1613 int dentalloc = 128;
1614 int dentoffs = 0;
1615 eio_ino_t inode_bits = 0;
1616#ifdef _WIN32
1617 HANDLE dirp;
1618 WIN32_FIND_DATA entp;
1619#else
713 DIR *dirp; 1620 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1622#endif
721 int memlen = 4096;
722 int memofs = 0;
723 int res = 0;
724 1623
725 X_LOCK (wrklock); 1624 req->result = -1;
1625
1626 if (!(flags & EIO_READDIR_DENTS))
1627 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1628
1629#ifdef _WIN32
1630 {
1631 int len = strlen ((const char *)req->ptr1);
1632 char *path = malloc (MAX_PATH);
1633 const char *fmt;
1634
1635 if (!len)
1636 fmt = "./*";
1637 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1638 fmt = "%s*";
1639 else
1640 fmt = "%s/*";
1641
1642 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1643 dirp = FindFirstFile (path, &entp);
1644 free (path);
1645
1646 if (dirp == INVALID_HANDLE_VALUE)
1647 {
1648 dirp = 0;
1649
1650 switch (GetLastError ())
1651 {
1652 case ERROR_FILE_NOT_FOUND:
1653 req->result = 0;
1654 break;
1655
1656 case ERROR_INVALID_NAME:
1657 case ERROR_PATH_NOT_FOUND:
1658 case ERROR_NO_MORE_FILES:
1659 errno = ENOENT;
1660 break;
1661
1662 case ERROR_NOT_ENOUGH_MEMORY:
1663 errno = ENOMEM;
1664 break;
1665
1666 default:
1667 errno = EINVAL;
1668 break;
1669 }
1670 }
1671 }
1672#else
726 self->dirp = dirp = opendir (req->ptr1); 1673 dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1674#endif
1675
728 req->flags |= EIO_FLAG_PTR2_FREE; 1676 if (req->flags & EIO_FLAG_PTR1_FREE)
1677 free (req->ptr1);
1678
1679 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1680 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1681 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock);
731 1682
732 if (dirp && u && names) 1683 if (dirp && names && (!flags || dents))
733 for (;;) 1684 for (;;)
734 { 1685 {
1686 int done;
1687
1688#ifdef _WIN32
1689 done = !dirp;
1690#else
735 errno = 0; 1691 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1692 entp = readdir (dirp);
1693 done = !entp;
1694#endif
737 1695
738 if (!entp) 1696 if (done)
1697 {
1698#ifndef _WIN32
1699 int old_errno = errno;
1700 closedir (dirp);
1701 errno = old_errno;
1702
1703 if (errno)
1704 break;
1705#endif
1706
1707 /* sort etc. */
1708 req->int1 = flags;
1709 req->result = dentoffs;
1710
1711 if (flags & EIO_READDIR_STAT_ORDER)
1712 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1713 else if (flags & EIO_READDIR_DIRS_FIRST)
1714 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1715 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1716 else
1717 {
1718 /* in this case, all is known, and we just put dirs first and sort them */
1719 eio_dirent *oth = dents + dentoffs;
1720 eio_dirent *dir = dents;
1721
1722 /* now partition dirs to the front, and non-dirs to the back */
1723 /* by walking from both sides and swapping if necessary */
1724 while (oth > dir)
1725 {
1726 if (dir->type == EIO_DT_DIR)
1727 ++dir;
1728 else if ((--oth)->type == EIO_DT_DIR)
1729 {
1730 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1731
1732 ++dir;
1733 }
1734 }
1735
1736 /* now sort the dirs only (dirs all have the same score) */
1737 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1738 }
1739
739 break; 1740 break;
1741 }
740 1742
741 name = entp->d_name; 1743 /* now add the entry to our list(s) */
1744 name = D_NAME (entp);
742 1745
1746 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1748 {
745 int len = strlen (name) + 1; 1749 int len = D_NAMLEN (entp) + 1;
746 1750
747 res++; 1751 while (ecb_expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1752 {
751 memlen *= 2; 1753 namesalloc *= 2;
752 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1754 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock);
755 1755
756 if (!names) 1756 if (!names)
757 break; 1757 break;
758 } 1758 }
759 1759
760 memcpy (names + memofs, name, len); 1760 memcpy (names + namesoffs, name, len);
1761
1762 if (dents)
1763 {
1764 struct eio_dirent *ent;
1765
1766 if (ecb_expect_false (dentoffs == dentalloc))
1767 {
1768 dentalloc *= 2;
1769 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1770
1771 if (!dents)
1772 break;
1773 }
1774
1775 ent = dents + dentoffs;
1776
1777 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1778 ent->namelen = len - 1;
1779 ent->inode = D_INO (entp);
1780
1781 inode_bits |= ent->inode;
1782
1783 switch (D_TYPE (entp))
1784 {
1785 default:
1786 ent->type = EIO_DT_UNKNOWN;
1787 flags |= EIO_READDIR_FOUND_UNKNOWN;
1788 break;
1789
1790 #ifdef DT_FIFO
1791 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1792 #endif
1793 #ifdef DT_CHR
1794 case DT_CHR: ent->type = EIO_DT_CHR; break;
1795 #endif
1796 #ifdef DT_MPC
1797 case DT_MPC: ent->type = EIO_DT_MPC; break;
1798 #endif
1799 #ifdef DT_DIR
1800 case DT_DIR: ent->type = EIO_DT_DIR; break;
1801 #endif
1802 #ifdef DT_NAM
1803 case DT_NAM: ent->type = EIO_DT_NAM; break;
1804 #endif
1805 #ifdef DT_BLK
1806 case DT_BLK: ent->type = EIO_DT_BLK; break;
1807 #endif
1808 #ifdef DT_MPB
1809 case DT_MPB: ent->type = EIO_DT_MPB; break;
1810 #endif
1811 #ifdef DT_REG
1812 case DT_REG: ent->type = EIO_DT_REG; break;
1813 #endif
1814 #ifdef DT_NWK
1815 case DT_NWK: ent->type = EIO_DT_NWK; break;
1816 #endif
1817 #ifdef DT_CMP
1818 case DT_CMP: ent->type = EIO_DT_CMP; break;
1819 #endif
1820 #ifdef DT_LNK
1821 case DT_LNK: ent->type = EIO_DT_LNK; break;
1822 #endif
1823 #ifdef DT_SOCK
1824 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1825 #endif
1826 #ifdef DT_DOOR
1827 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1828 #endif
1829 #ifdef DT_WHT
1830 case DT_WHT: ent->type = EIO_DT_WHT; break;
1831 #endif
1832 }
1833
1834 ent->score = 7;
1835
1836 if (flags & EIO_READDIR_DIRS_FIRST)
1837 {
1838 if (ent->type == EIO_DT_UNKNOWN)
1839 {
1840 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1841 ent->score = 1;
1842 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1843 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1844 }
1845 else if (ent->type == EIO_DT_DIR)
1846 ent->score = 0;
1847 }
1848 }
1849
761 memofs += len; 1850 namesoffs += len;
1851 ++dentoffs;
762 } 1852 }
1853
1854 if (EIO_CANCELLED (req))
1855 {
1856 errno = ECANCELED;
1857 break;
1858 }
1859
1860#ifdef _WIN32
1861 if (!FindNextFile (dirp, &entp))
1862 {
1863 FindClose (dirp);
1864 dirp = 0;
1865 }
1866#endif
763 } 1867 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769} 1868}
770 1869
771/*****************************************************************************/ 1870/*****************************************************************************/
772 1871
1872#define ALLOC(len) \
1873 if (!req->ptr2) \
1874 { \
1875 X_LOCK (wrklock); \
1876 req->flags |= EIO_FLAG_PTR2_FREE; \
1877 X_UNLOCK (wrklock); \
1878 req->ptr2 = malloc (len); \
1879 if (!req->ptr2) \
1880 { \
1881 errno = ENOMEM; \
1882 req->result = -1; \
1883 break; \
1884 } \
1885 }
1886
773X_THREAD_PROC (eio_proc) 1887X_THREAD_PROC (etp_proc)
774{ 1888{
775 eio_req *req; 1889 ETP_REQ *req;
776 struct timespec ts; 1890 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1891 etp_worker *self = (etp_worker *)thr_arg;
778 1892
779 /* try to distribute timeouts somewhat randomly */ 1893 /* try to distribute timeouts somewhat evenly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1894 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1895
782 for (;;) 1896 for (;;)
783 { 1897 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1898 ts.tv_sec = 0;
785 1899
786 X_LOCK (reqlock); 1900 X_LOCK (reqlock);
787 1901
788 for (;;) 1902 for (;;)
789 { 1903 {
790 self->req = req = reqq_shift (&req_queue); 1904 self->req = req = reqq_shift (&req_queue);
791 1905
792 if (req) 1906 if (req)
793 break; 1907 break;
794 1908
1909 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1910 {
1911 X_UNLOCK (reqlock);
1912 X_LOCK (wrklock);
1913 --started;
1914 X_UNLOCK (wrklock);
1915 goto quit;
1916 }
1917
795 ++idle; 1918 ++idle;
796 1919
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1920 if (idle <= max_idle)
1921 /* we are allowed to idle, so do so without any timeout */
1922 X_COND_WAIT (reqwait, reqlock);
1923 else
798 { 1924 {
799 if (idle > max_idle) 1925 /* initialise timeout once */
800 { 1926 if (!ts.tv_sec)
801 --idle; 1927 ts.tv_sec = time (0) + idle_timeout;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808 1928
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1929 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
811 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1930 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
812 } 1931 }
813 1932
814 --idle; 1933 --idle;
815 } 1934 }
816 1935
817 --nready; 1936 --nready;
818 1937
819 X_UNLOCK (reqlock); 1938 X_UNLOCK (reqlock);
820 1939
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type) 1940 if (req->type < 0)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit; 1941 goto quit;
910 1942
911 default: 1943 ETP_EXECUTE (self, req);
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1944
918 X_LOCK (reslock); 1945 X_LOCK (reslock);
919 1946
920 ++npending; 1947 ++npending;
921 1948
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1949 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1950 want_poll_cb ();
924 1951
925 self->req = 0; 1952 self->req = 0;
926 worker_clear (self); 1953 etp_worker_clear (self);
927 1954
928 X_UNLOCK (reslock); 1955 X_UNLOCK (reslock);
929 } 1956 }
930 1957
931quit: 1958quit:
932 X_LOCK (wrklock); 1959 X_LOCK (wrklock);
933 worker_free (self); 1960 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1961 X_UNLOCK (wrklock);
935 1962
936 return 0; 1963 return 0;
937} 1964}
938 1965
939/*****************************************************************************/ 1966/*****************************************************************************/
940 1967
941static void eio_atfork_prepare (void) 1968int ecb_cold
1969eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1970{
943 X_LOCK (wrklock);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE 1971#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952}
953
954static void eio_atfork_parent (void)
955{
956#if !HAVE_READDIR_R
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock); 1972 X_MUTEX_CREATE (preadwritelock);
961#endif 1973#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965}
966 1974
967static void eio_atfork_child (void) 1975 return etp_init (want_poll, done_poll);
968{ 1976}
969 eio_req *prv;
970 1977
971 while (prv = reqq_shift (&req_queue)) 1978ecb_inline void
972 eio_destroy (prv); 1979eio_api_destroy (eio_req *req)
1980{
1981 free (req);
1982}
973 1983
974 while (prv = reqq_shift (&res_queue)) 1984#define REQ(rtype) \
975 eio_destroy (prv); 1985 eio_req *req; \
1986 \
1987 req = (eio_req *)calloc (1, sizeof *req); \
1988 if (!req) \
1989 return 0; \
1990 \
1991 req->type = rtype; \
1992 req->pri = pri; \
1993 req->finish = cb; \
1994 req->data = data; \
1995 req->destroy = eio_api_destroy;
976 1996
977 while (wrk_first.next != &wrk_first) 1997#define SEND eio_submit (req); return req
1998
1999#define PATH \
2000 req->flags |= EIO_FLAG_PTR1_FREE; \
2001 req->ptr1 = strdup (path); \
2002 if (!req->ptr1) \
2003 { \
2004 eio_api_destroy (req); \
2005 return 0; \
2006 }
2007
2008static void
2009eio_execute (etp_worker *self, eio_req *req)
2010{
2011 if (ecb_expect_false (EIO_CANCELLED (req)))
978 { 2012 {
979 worker *wrk = wrk_first.next; 2013 req->result = -1;
980 2014 req->errorno = ECANCELED;
981 if (wrk->req) 2015 return;
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 } 2016 }
987 2017
988 started = 0; 2018 switch (req->type)
989 idle = 0; 2019 {
990 nreqs = 0; 2020 case EIO_READ: ALLOC (req->size);
991 nready = 0; 2021 req->result = req->offs >= 0
992 npending = 0; 2022 ? pread (req->int1, req->ptr2, req->size, req->offs)
2023 : read (req->int1, req->ptr2, req->size); break;
2024 case EIO_WRITE: req->result = req->offs >= 0
2025 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2026 : write (req->int1, req->ptr2, req->size); break;
993 2027
994 eio_atfork_parent (); 2028 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
995} 2029 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
996 2030
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2031 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
998{ 2032 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
999 want_poll_cb = want_poll; 2033 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1000 done_poll_cb = done_poll; 2034 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2035 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2036 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1001 2037
2038 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2039 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2040 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2041 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2042
2043 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2044 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2045 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2046 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2047 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2048 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2049
2050 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2051 case EIO_CLOSE: req->result = close (req->int1); break;
2052 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2053 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2054 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2055 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2056 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2057 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2058 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2059 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2060
2061 case EIO_REALPATH: eio__realpath (req, self); break;
2062
2063 case EIO_READLINK: ALLOC (PATH_MAX);
2064 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2065
2066 case EIO_SYNC: req->result = 0; sync (); break;
2067 case EIO_FSYNC: req->result = fsync (req->int1); break;
2068 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2069 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2070 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2071 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2072 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2073 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2074 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2075
2076 case EIO_READDIR: eio__scandir (req, self); break;
2077
2078 case EIO_BUSY:
1002#ifdef _WIN32 2079#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 2080 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 2081#else
1005 X_MUTEX_CHECK (reqlock); 2082 {
1006 X_MUTEX_CHECK (reqwait); 2083 struct timeval tv;
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009 2084
1010 X_COND_CHECK (reqwait); 2085 tv.tv_sec = req->nv1;
1011#endif 2086 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012 2087
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2088 req->result = select (0, 0, 0, 0, &tv);
1014} 2089 }
2090#endif
2091 break;
1015 2092
1016#if 0 2093 case EIO_UTIME:
2094 case EIO_FUTIME:
2095 {
2096 struct timeval tv[2];
2097 struct timeval *times;
1017 2098
2099 if (req->nv1 != -1. || req->nv2 != -1.)
2100 {
2101 tv[0].tv_sec = req->nv1;
2102 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2103 tv[1].tv_sec = req->nv2;
2104 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2105
2106 times = tv;
2107 }
2108 else
2109 times = 0;
2110
2111 req->result = req->type == EIO_FUTIME
2112 ? futimes (req->int1, times)
2113 : utimes (req->ptr1, times);
2114 }
2115 break;
2116
2117 case EIO_GROUP:
2118 abort (); /* handled in eio_request */
2119
2120 case EIO_NOP:
2121 req->result = 0;
2122 break;
2123
2124 case EIO_CUSTOM:
2125 req->feed (req);
2126 break;
2127
2128 default:
2129 errno = ENOSYS;
2130 req->result = -1;
2131 break;
2132 }
2133
2134 req->errorno = errno;
2135}
2136
2137#ifndef EIO_NO_WRAPPERS
2138
2139eio_req *eio_nop (int pri, eio_cb cb, void *data)
2140{
2141 REQ (EIO_NOP); SEND;
2142}
2143
2144eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2145{
2146 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2147}
2148
1018eio_req *eio_fsync (int fd, eio_cb cb); 2149eio_req *eio_sync (int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_SYNC); SEND;
2152}
2153
2154eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2155{
2156 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2157}
2158
2159eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2160{
2161 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2162}
2163
2164eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2167}
2168
2169eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2170{
2171 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2172}
2173
2174eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2175{
2176 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2177}
2178
2179eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2182}
2183
2184eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2187}
2188
1019eio_req *eio_fdatasync (int fd, eio_cb cb); 2189eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020eio_req *eio_dupclose (int fd, eio_cb cb); 2190{
2191 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2192}
2193
2194eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2195{
2196 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2197}
2198
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2199eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2200{
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2201 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2202}
2203
2204eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2205{
2206 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2207}
2208
2209eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2210{
2211 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2212}
2213
2214eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2215{
2216 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2217}
2218
2219eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2220{
2221 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2222}
2223
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2224eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2225{
2226 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2227}
2228
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2229eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2230{
2231 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2232}
2233
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2234eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2235{
2236 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2237}
2238
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2239eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2240{
2241 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2242}
2243
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2244eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2245{
2246 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2247}
2248
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2249eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2250{
2251 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2252}
2253
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2254eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2255{
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2256 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2257}
2258
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2259eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2260{
2261 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2262}
2263
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2264eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2265{
2266 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2267}
2268
2269eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2270{
2271 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2272}
2273
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2274eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2275{
2276 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2277}
2278
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2279eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2280{
2281 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2282}
2283
2284static eio_req *
2285eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2286{
2287 REQ (type); PATH; SEND;
2288}
2289
2290eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2291{
2292 return eio__1path (EIO_READLINK, path, pri, cb, data);
2293}
2294
2295eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2296{
2297 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2298}
2299
2300eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2301{
2302 return eio__1path (EIO_STAT, path, pri, cb, data);
2303}
2304
2305eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2306{
2307 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2308}
2309
2310eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2311{
2312 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2313}
2314
1040eio_req *eio_unlink (const char *path, eio_cb cb); 2315eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2316{
2317 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2318}
2319
1041eio_req *eio_rmdir (const char *path, eio_cb cb); 2320eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2321{
2322 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2323}
2324
2325eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2326{
2327 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2328}
2329
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2330eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2331{
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2332 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2333}
2334
2335static eio_req *
2336eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2337{
2338 REQ (type); PATH;
2339
2340 req->flags |= EIO_FLAG_PTR2_FREE;
2341 req->ptr2 = strdup (new_path);
2342 if (!req->ptr2)
2343 {
2344 eio_api_destroy (req);
2345 return 0;
2346 }
2347
2348 SEND;
2349}
2350
2351eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2352{
2353 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2354}
2355
2356eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2357{
2358 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2359}
2360
2361eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2362{
2363 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2364}
2365
2366eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2367{
2368 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2369}
2370
2371#endif
2372
2373eio_req *eio_grp (eio_cb cb, void *data)
2374{
2375 const int pri = EIO_PRI_MAX;
2376
2377 REQ (EIO_GROUP); SEND;
2378}
2379
2380#undef REQ
2381#undef PATH
2382#undef SEND
2383
2384/*****************************************************************************/
2385/* grp functions */
2386
1046void 2387void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 {
1130 dREQ;
1131
1132 req->type = ix;
1133 req->sv1 = newSVsv (fh);
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141
1142 if (!SvREADONLY (data))
1143 {
1144 SvREADONLY_on (data);
1145 req->flags |= FLAG_SV2_RO_OFF;
1146 }
1147
1148 EIO_SEND;
1149 }
1150}
1151
1152void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159
1160 data = newSV (NAME_MAX);
1161 SvPOK_on (data);
1162
1163 req->type = EIO_READLINK;
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170}
1171
1172void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2388eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2389{
1421 grp->int2 = limit; 2390 grp->int2 = limit;
1422 grp->feed = feed; 2391 grp->feed = feed;
1423 2392
1424 grp_try_feed (grp); 2393 grp_try_feed (grp);
1425} 2394}
1426 2395
2396void
1427void eio_grp_limit (eio_req *grp, int limit) 2397eio_grp_limit (eio_req *grp, int limit)
1428{ 2398{
1429 grp->int2 = limit; 2399 grp->int2 = limit;
1430 2400
1431 grp_try_feed (grp); 2401 grp_try_feed (grp);
1432} 2402}
1433 2403
2404void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2405eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2406{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2408
2409 grp->flags |= EIO_FLAG_GROUPADD;
1437 2410
1438 ++grp->size; 2411 ++grp->size;
1439 req->grp = grp; 2412 req->grp = grp;
1440 2413
1441 req->grp_prev = 0; 2414 req->grp_prev = 0;
1445 grp->grp_first->grp_prev = req; 2418 grp->grp_first->grp_prev = req;
1446 2419
1447 grp->grp_first = req; 2420 grp->grp_first = req;
1448} 2421}
1449 2422
2423/*****************************************************************************/
2424/* misc garbage */
1450 2425
2426eio_ssize_t
2427eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2428{
2429 return eio__sendfile (ofd, ifd, offset, count);
2430}
2431

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines