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

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
… …
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#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);
81
15#ifndef EIO_FINISH 82#ifndef EIO_FINISH
16# 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
17#endif 84#endif
18 85
19#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
22 89
23#ifndef EIO_FEED 90#ifndef EIO_FEED
24# 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)
25#endif 92#endif
26 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
27#ifdef _WIN32 105#ifdef _WIN32
28 106
29 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 196
31#else 197#else
32 198
33# include "config.h"
34# include <sys/time.h> 199 #include <sys/time.h>
35# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
36# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
37# include <utime.h> 237# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 238#endif
44 239
240#if HAVE_SYS_SYSCALL_H
241# include <sys/syscall.h>
242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
45#endif 246#endif
46 247
47#if HAVE_SENDFILE 248#if HAVE_SENDFILE
48# if __linux 249# if __linux
49# include <sys/sendfile.h> 250# include <sys/sendfile.h>
50# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 252# include <sys/socket.h>
52# include <sys/uio.h> 253# include <sys/uio.h>
53# elif __hpux 254# elif __hpux
54# include <sys/socket.h> 255# include <sys/socket.h>
55# elif __solaris /* not yet */ 256# elif __solaris
56# include <sys/sendfile.h> 257# include <sys/sendfile.h>
57# else 258# else
58# error sendfile support requested but not available 259# error sendfile support requested but not available
59# endif 260# endif
60#endif 261#endif
61 262
62/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
64 272
65/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 274#ifndef NAME_MAX
67# define NAME_MAX 4096 275# define NAME_MAX 4096
68#endif 276#endif
69 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
70/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
72 285
73#define dBUF \ 286#define dBUF \
74 char *eio_buf; \
75 X_LOCK (etplock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (etplock); \
78 errno = ENOMEM; \ 288 errno = ENOMEM; \
79 if (!eio_buf) \ 289 if (!eio_buf) \
80 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
81 294
82#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
83 296
297#define ETP_PRI_MIN EIO_PRI_MIN
298#define ETP_PRI_MAX EIO_PRI_MAX
299
300struct etp_worker;
301
302#define ETP_REQ eio_req
303#define ETP_DESTROY(req) eio_destroy (req)
304static int eio_finish (eio_req *req);
305#define ETP_FINISH(req) eio_finish (req)
306static void eio_execute (struct etp_worker *self, eio_req *req);
307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
308
84/*****************************************************************************/ 309/*****************************************************************************/
85 310
311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
312
86/* calculcate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
314ecb_inline int
87static int tvdiff (struct timeval *tv1, struct timeval *tv2) 315tvdiff (struct timeval *tv1, struct timeval *tv2)
88{ 316{
89 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 317 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 318 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
91} 319}
92 320
93static unsigned int started, idle, wanted = 4; 321static unsigned int started, idle, wanted = 4;
94 322
95typedef struct etp_pool
96{
97 void (*want_poll_cb) (void); 323static void (*want_poll_cb) (void);
98 void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
99 325
100 unsigned int max_poll_time; 326static unsigned int max_poll_time; /* reslock */
101 unsigned int max_poll_reqs; 327static unsigned int max_poll_reqs; /* reslock */
102} etp_pool;
103 328
104static volatile unsigned int nreqs, nready, npending; 329static unsigned int nreqs; /* reqlock */
105static volatile unsigned int max_idle = 4; 330static unsigned int nready; /* reqlock */
331static unsigned int npending; /* reqlock */
332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
106 334
107static mutex_t etplock = X_MUTEX_INIT; 335static xmutex_t wrklock;
108static mutex_t reslock = X_MUTEX_INIT; 336static xmutex_t reslock;
109static mutex_t reqlock = X_MUTEX_INIT; 337static xmutex_t reqlock;
110static cond_t reqwait = X_COND_INIT; 338static xcond_t reqwait;
111 339
340#if !HAVE_PREADWRITE
341/*
342 * make our pread/pwrite emulation safe against themselves, but not against
343 * normal read/write by using a mutex. slows down execution a lot,
344 * but that's your problem, not mine.
345 */
346static xmutex_t preadwritelock;
347#endif
348
112typedef struct worker 349typedef struct etp_worker
113{ 350{
114 /* locked by etplock */ 351 /* locked by wrklock */
115 struct worker *prev, *next; 352 struct etp_worker *prev, *next;
116 353
117 thread_t tid; 354 xthread_t tid;
118 355
119 /* locked by reslock, reqlock or etplock */ 356 /* locked by reslock, reqlock or wrklock */
120 eio_req *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
121 void *dbuf; 358
122 DIR *dirp; 359#ifdef ETP_WORKER_COMMON
360 ETP_WORKER_COMMON
361#endif
123} worker; 362} etp_worker;
124 363
125static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
365
366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126 368
127/* worker threads management */ 369/* worker threads management */
128 370
371static void ecb_cold
129static void worker_clear (worker *wrk) 372etp_worker_clear (etp_worker *wrk)
130{ 373{
131 if (wrk->dirp)
132 {
133 closedir (wrk->dirp);
134 wrk->dirp = 0;
135 }
136
137 if (wrk->dbuf)
138 {
139 free (wrk->dbuf);
140 wrk->dbuf = 0;
141 }
142} 374}
143 375
144static void worker_free (worker *wrk) 376static void ecb_cold
377etp_worker_free (etp_worker *wrk)
145{ 378{
146 wrk->next->prev = wrk->prev; 379 wrk->next->prev = wrk->prev;
147 wrk->prev->next = wrk->next; 380 wrk->prev->next = wrk->next;
148 381
149 free (wrk); 382 free (wrk);
150} 383}
151 384
152unsigned int eio_nreqs (void) 385static unsigned int
386etp_nreqs (void)
153{ 387{
388 int retval;
389 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
390 retval = nreqs;
391 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 392 return retval;
155} 393}
156 394
157unsigned int eio_nready (void) 395static unsigned int
396etp_nready (void)
158{ 397{
159 unsigned int retval; 398 unsigned int retval;
160 399
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 401 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 403
165 return retval; 404 return retval;
166} 405}
167 406
168unsigned int eio_npending (void) 407static unsigned int
408etp_npending (void)
169{ 409{
170 unsigned int retval; 410 unsigned int retval;
171 411
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 413 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 415
176 return retval; 416 return retval;
177} 417}
178 418
179unsigned int eio_nthreads (void) 419static unsigned int
420etp_nthreads (void)
180{ 421{
181 unsigned int retval; 422 unsigned int retval;
182 423
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 424 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 425 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 432 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 433 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 434 * per shift, the most expensive operation.
194 */ 435 */
195typedef struct { 436typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 437 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 438 int size;
198} reqq; 439} etp_reqq;
199 440
200static reqq req_queue; 441static etp_reqq req_queue;
201static reqq res_queue; 442static etp_reqq res_queue;
202 443
203static int reqq_push (reqq *q, eio_req *req) 444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
455static int ecb_noinline
456reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 457{
205 int pri = req->pri; 458 int pri = req->pri;
206 req->next = 0; 459 req->next = 0;
207 460
208 if (q->qe[pri]) 461 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 467 q->qe[pri] = q->qs[pri] = req;
215 468
216 return q->size++; 469 return q->size++;
217} 470}
218 471
219static eio_req *reqq_shift (reqq *q) 472static ETP_REQ * ecb_noinline
473reqq_shift (etp_reqq *q)
220{ 474{
221 int pri; 475 int pri;
222 476
223 if (!q->size) 477 if (!q->size)
224 return 0; 478 return 0;
225 479
226 --q->size; 480 --q->size;
227 481
228 for (pri = EIO_NUM_PRI; pri--; ) 482 for (pri = ETP_NUM_PRI; pri--; )
229 { 483 {
230 eio_req *req = q->qs[pri]; 484 eio_req *req = q->qs[pri];
231 485
232 if (req) 486 if (req)
233 { 487 {
… …
239 } 493 }
240 494
241 abort (); 495 abort ();
242} 496}
243 497
244static void etp_atfork_prepare (void) 498static int ecb_cold
499etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 500{
246 X_LOCK (etplock); 501 X_MUTEX_CREATE (wrklock);
247 X_LOCK (reqlock); 502 X_MUTEX_CREATE (reslock);
248 X_LOCK (reslock); 503 X_MUTEX_CREATE (reqlock);
249#if !HAVE_PREADWRITE 504 X_COND_CREATE (reqwait);
250 X_LOCK (preadwritelock);
251#endif
252#if !HAVE_READDIR_R
253 X_LOCK (readdirlock);
254#endif
255}
256 505
257static void etp_atfork_parent (void) 506 reqq_init (&req_queue);
258{ 507 reqq_init (&res_queue);
259#if !HAVE_READDIR_R
260 X_UNLOCK (readdirlock);
261#endif
262#if !HAVE_PREADWRITE
263 X_UNLOCK (preadwritelock);
264#endif
265 X_UNLOCK (reslock);
266 X_UNLOCK (reqlock);
267 X_UNLOCK (etplock);
268}
269 508
270static void etp_atfork_child (void) 509 wrk_first.next =
271{ 510 wrk_first.prev = &wrk_first;
272 eio_req *prv;
273
274 while (prv = reqq_shift (&req_queue))
275 eio_destroy (prv);
276
277 while (prv = reqq_shift (&res_queue))
278 eio_destroy (prv);
279
280 while (wrk_first.next != &wrk_first)
281 {
282 worker *wrk = wrk_first.next;
283
284 if (wrk->req)
285 eio_destroy (wrk->req);
286
287 worker_clear (wrk);
288 worker_free (wrk);
289 }
290 511
291 started = 0; 512 started = 0;
292 idle = 0; 513 idle = 0;
293 nreqs = 0; 514 nreqs = 0;
294 nready = 0; 515 nready = 0;
295 npending = 0; 516 npending = 0;
296 517
297 etp_atfork_parent ();
298}
299
300static void
301etp_once_init (void)
302{
303 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304}
305
306static int
307etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))
308{
309 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310
311 pthread_once (&doinit, etp_once_init);
312
313 memset (etp, 0, sizeof *etp);
314
315 etp->want_poll_cb = want_poll; 518 want_poll_cb = want_poll;
316 etp->done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
317}
318 520
319static etp_pool etp;
320
321/*****************************************************************************/
322
323static void grp_try_feed (eio_req *grp)
324{
325 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
326 {
327 int old_len = grp->size;
328
329 EIO_FEED (grp);
330
331 /* stop if no progress has been made */
332 if (old_len == grp->size)
333 {
334 grp->feed = 0;
335 break;
336 }
337 }
338}
339
340static int eio_finish (eio_req *req);
341
342static int grp_dec (eio_req *grp)
343{
344 --grp->size;
345
346 /* call feeder, if applicable */
347 grp_try_feed (grp);
348
349 /* finish, if done */
350 if (!grp->size && grp->int1)
351 return eio_finish (grp);
352 else
353 return 0; 521 return 0;
354} 522}
355 523
356void eio_destroy (eio_req *req)
357{
358 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
359 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
360
361 EIO_DESTROY (req);
362}
363
364static int eio_finish (eio_req *req)
365{
366 int res = EIO_FINISH (req);
367
368 if (req->grp)
369 {
370 int res2;
371 eio_req *grp = req->grp;
372
373 /* unlink request */
374 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
375 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
376
377 if (grp->grp_first == req)
378 grp->grp_first = req->grp_next;
379
380 res2 = grp_dec (grp);
381
382 if (!res && res2)
383 res = res2;
384 }
385
386 eio_destroy (req);
387
388 return res;
389}
390
391void eio_grp_cancel (eio_req *grp)
392{
393 for (grp = grp->grp_first; grp; grp = grp->grp_next)
394 eio_cancel (grp);
395}
396
397void eio_cancel (eio_req *req)
398{
399 X_LOCK (etplock);
400 req->flags |= EIO_FLAG_CANCELLED;
401 X_UNLOCK (etplock);
402
403 eio_grp_cancel (req);
404}
405
406X_THREAD_PROC (eio_proc); 524X_THREAD_PROC (etp_proc);
407 525
408static void start_thread (void) 526static void ecb_cold
527etp_start_thread (void)
409{ 528{
410 worker *wrk = calloc (1, sizeof (worker)); 529 etp_worker *wrk = calloc (1, sizeof (etp_worker));
411 530
412 /*TODO*/ 531 /*TODO*/
413 assert (("unable to allocate worker thread data", wrk)); 532 assert (("unable to allocate worker thread data", wrk));
414 533
415 X_LOCK (etplock); 534 X_LOCK (wrklock);
416 535
417 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 536 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418 { 537 {
419 wrk->prev = &wrk_first; 538 wrk->prev = &wrk_first;
420 wrk->next = wrk_first.next; 539 wrk->next = wrk_first.next;
421 wrk_first.next->prev = wrk; 540 wrk_first.next->prev = wrk;
422 wrk_first.next = wrk; 541 wrk_first.next = wrk;
423 ++started; 542 ++started;
424 } 543 }
425 else 544 else
426 free (wrk); 545 free (wrk);
427 546
428 X_UNLOCK (etplock); 547 X_UNLOCK (wrklock);
429} 548}
430 549
550static void
431static void maybe_start_thread (void) 551etp_maybe_start_thread (void)
432{ 552{
433 if (eio_nthreads () >= wanted) 553 if (ecb_expect_true (etp_nthreads () >= wanted))
434 return; 554 return;
435 555
436 /* todo: maybe use idle here, but might be less exact */ 556 /* todo: maybe use idle here, but might be less exact */
437 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 557 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438 return; 558 return;
439 559
440 start_thread (); 560 etp_start_thread ();
441} 561}
442 562
443void eio_submit (eio_req *req) 563static void ecb_cold
444{ 564etp_end_thread (void)
445 req->pri += EIO_PRI_BIAS;
446
447 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
448 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
449
450 ++nreqs;
451
452 X_LOCK (reqlock);
453 ++nready;
454 reqq_push (&req_queue, req);
455 X_COND_SIGNAL (reqwait);
456 X_UNLOCK (reqlock);
457
458 maybe_start_thread ();
459}
460
461static void end_thread (void)
462{ 565{
463 eio_req *req = calloc (1, sizeof (eio_req)); 566 eio_req *req = calloc (1, sizeof (eio_req));
464 567
465 req->type = EIO_QUIT; 568 req->type = -1;
466 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 569 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467 570
468 X_LOCK (reqlock); 571 X_LOCK (reqlock);
469 reqq_push (&req_queue, req); 572 reqq_push (&req_queue, req);
470 X_COND_SIGNAL (reqwait); 573 X_COND_SIGNAL (reqwait);
471 X_UNLOCK (reqlock); 574 X_UNLOCK (reqlock);
472 575
473 X_LOCK (etplock); 576 X_LOCK (wrklock);
474 --started; 577 --started;
475 X_UNLOCK (etplock); 578 X_UNLOCK (wrklock);
476} 579}
477 580
478void eio_set_max_poll_time (double nseconds) 581static int
479{
480 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
481 etp.max_poll_time = nseconds;
482 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
483}
484
485void eio_set_max_poll_reqs (unsigned int maxreqs)
486{
487 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
488 etp.max_poll_reqs = maxreqs;
489 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
490}
491
492void eio_set_max_idle (unsigned int nthreads)
493{
494 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
495 max_idle = nthreads <= 0 ? 1 : nthreads;
496 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
497}
498
499void eio_set_min_parallel (unsigned int nthreads)
500{
501 if (wanted < nthreads)
502 wanted = nthreads;
503}
504
505void eio_set_max_parallel (unsigned int nthreads)
506{
507 if (wanted > nthreads)
508 wanted = nthreads;
509
510 while (started > wanted)
511 end_thread ();
512}
513
514int eio_poll (void) 582etp_poll (void)
515{ 583{
516 int maxreqs = etp.max_poll_reqs; 584 unsigned int maxreqs;
585 unsigned int maxtime;
517 struct timeval tv_start, tv_now; 586 struct timeval tv_start, tv_now;
518 eio_req *req;
519 587
520 if (etp.max_poll_time) 588 X_LOCK (reslock);
589 maxreqs = max_poll_reqs;
590 maxtime = max_poll_time;
591 X_UNLOCK (reslock);
592
593 if (maxtime)
521 gettimeofday (&tv_start, 0); 594 gettimeofday (&tv_start, 0);
522 595
523 for (;;) 596 for (;;)
524 { 597 {
598 ETP_REQ *req;
599
525 maybe_start_thread (); 600 etp_maybe_start_thread ();
526 601
527 X_LOCK (reslock); 602 X_LOCK (reslock);
528 req = reqq_shift (&res_queue); 603 req = reqq_shift (&res_queue);
529 604
530 if (req) 605 if (req)
531 { 606 {
532 --npending; 607 --npending;
533 608
534 if (!res_queue.size && etp.done_poll_cb) 609 if (!res_queue.size && done_poll_cb)
535 etp.done_poll_cb (); 610 done_poll_cb ();
536 } 611 }
537 612
538 X_UNLOCK (reslock); 613 X_UNLOCK (reslock);
539 614
540 if (!req) 615 if (!req)
541 return 0; 616 return 0;
542 617
618 X_LOCK (reqlock);
543 --nreqs; 619 --nreqs;
620 X_UNLOCK (reqlock);
544 621
545 if (req->type == EIO_GROUP && req->size) 622 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
546 { 623 {
547 req->int1 = 1; /* mark request as delayed */ 624 req->int1 = 1; /* mark request as delayed */
548 continue; 625 continue;
549 } 626 }
550 else 627 else
551 { 628 {
552 int res = eio_finish (req); 629 int res = ETP_FINISH (req);
553 if (res) 630 if (ecb_expect_false (res))
554 return res; 631 return res;
555 } 632 }
556 633
557 if (maxreqs && !--maxreqs) 634 if (ecb_expect_false (maxreqs && !--maxreqs))
558 break; 635 break;
559 636
560 if (etp.max_poll_time) 637 if (maxtime)
561 { 638 {
562 gettimeofday (&tv_now, 0); 639 gettimeofday (&tv_now, 0);
563 640
564 if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time) 641 if (tvdiff (&tv_start, &tv_now) >= maxtime)
565 break; 642 break;
566 } 643 }
567 } 644 }
568 645
569 errno = EAGAIN; 646 errno = EAGAIN;
570 return -1; 647 return -1;
571} 648}
572 649
650static void
651etp_cancel (ETP_REQ *req)
652{
653 req->cancelled = 1;
654
655 eio_grp_cancel (req);
656}
657
658static void
659etp_submit (ETP_REQ *req)
660{
661 req->pri -= ETP_PRI_MIN;
662
663 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
664 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
665
666 if (ecb_expect_false (req->type == EIO_GROUP))
667 {
668 /* I hope this is worth it :/ */
669 X_LOCK (reqlock);
670 ++nreqs;
671 X_UNLOCK (reqlock);
672
673 X_LOCK (reslock);
674
675 ++npending;
676
677 if (!reqq_push (&res_queue, req) && want_poll_cb)
678 want_poll_cb ();
679
680 X_UNLOCK (reslock);
681 }
682 else
683 {
684 X_LOCK (reqlock);
685 ++nreqs;
686 ++nready;
687 reqq_push (&req_queue, req);
688 X_COND_SIGNAL (reqwait);
689 X_UNLOCK (reqlock);
690
691 etp_maybe_start_thread ();
692 }
693}
694
695static void ecb_cold
696etp_set_max_poll_time (double nseconds)
697{
698 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
699 max_poll_time = nseconds * EIO_TICKS;
700 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
701}
702
703static void ecb_cold
704etp_set_max_poll_reqs (unsigned int maxreqs)
705{
706 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
707 max_poll_reqs = maxreqs;
708 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
709}
710
711static void ecb_cold
712etp_set_max_idle (unsigned int nthreads)
713{
714 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
715 max_idle = nthreads;
716 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
717}
718
719static void ecb_cold
720etp_set_idle_timeout (unsigned int seconds)
721{
722 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
723 idle_timeout = seconds;
724 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
725}
726
727static void ecb_cold
728etp_set_min_parallel (unsigned int nthreads)
729{
730 if (wanted < nthreads)
731 wanted = nthreads;
732}
733
734static void ecb_cold
735etp_set_max_parallel (unsigned int nthreads)
736{
737 if (wanted > nthreads)
738 wanted = nthreads;
739
740 while (started > wanted)
741 etp_end_thread ();
742}
743
744/*****************************************************************************/
745
746static void
747grp_try_feed (eio_req *grp)
748{
749 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
750 {
751 grp->flags &= ~EIO_FLAG_GROUPADD;
752
753 EIO_FEED (grp);
754
755 /* stop if no progress has been made */
756 if (!(grp->flags & EIO_FLAG_GROUPADD))
757 {
758 grp->feed = 0;
759 break;
760 }
761 }
762}
763
764static int
765grp_dec (eio_req *grp)
766{
767 --grp->size;
768
769 /* call feeder, if applicable */
770 grp_try_feed (grp);
771
772 /* finish, if done */
773 if (!grp->size && grp->int1)
774 return eio_finish (grp);
775 else
776 return 0;
777}
778
779static void
780eio_destroy (eio_req *req)
781{
782 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
783 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
784
785 EIO_DESTROY (req);
786}
787
788static int
789eio_finish (eio_req *req)
790{
791 int res = EIO_FINISH (req);
792
793 if (req->grp)
794 {
795 int res2;
796 eio_req *grp = req->grp;
797
798 /* unlink request */
799 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
800 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
801
802 if (grp->grp_first == req)
803 grp->grp_first = req->grp_next;
804
805 res2 = grp_dec (grp);
806
807 if (!res)
808 res = res2;
809 }
810
811 eio_destroy (req);
812
813 return res;
814}
815
816void
817eio_grp_cancel (eio_req *grp)
818{
819 for (grp = grp->grp_first; grp; grp = grp->grp_next)
820 eio_cancel (grp);
821}
822
823void
824eio_cancel (eio_req *req)
825{
826 etp_cancel (req);
827}
828
829void
830eio_submit (eio_req *req)
831{
832 etp_submit (req);
833}
834
835unsigned int
836eio_nreqs (void)
837{
838 return etp_nreqs ();
839}
840
841unsigned int
842eio_nready (void)
843{
844 return etp_nready ();
845}
846
847unsigned int
848eio_npending (void)
849{
850 return etp_npending ();
851}
852
853unsigned int ecb_cold
854eio_nthreads (void)
855{
856 return etp_nthreads ();
857}
858
859void ecb_cold
860eio_set_max_poll_time (double nseconds)
861{
862 etp_set_max_poll_time (nseconds);
863}
864
865void ecb_cold
866eio_set_max_poll_reqs (unsigned int maxreqs)
867{
868 etp_set_max_poll_reqs (maxreqs);
869}
870
871void ecb_cold
872eio_set_max_idle (unsigned int nthreads)
873{
874 etp_set_max_idle (nthreads);
875}
876
877void ecb_cold
878eio_set_idle_timeout (unsigned int seconds)
879{
880 etp_set_idle_timeout (seconds);
881}
882
883void ecb_cold
884eio_set_min_parallel (unsigned int nthreads)
885{
886 etp_set_min_parallel (nthreads);
887}
888
889void ecb_cold
890eio_set_max_parallel (unsigned int nthreads)
891{
892 etp_set_max_parallel (nthreads);
893}
894
895int eio_poll (void)
896{
897 return etp_poll ();
898}
899
573/*****************************************************************************/ 900/*****************************************************************************/
574/* work around various missing functions */ 901/* work around various missing functions */
575 902
576#if !HAVE_PREADWRITE 903#if !HAVE_PREADWRITE
904# undef pread
905# undef pwrite
577# define pread eio__pread 906# define pread eio__pread
578# define pwrite eio__pwrite 907# define pwrite eio__pwrite
579 908
580/*
581 * make our pread/pwrite safe against themselves, but not against
582 * normal read/write by using a mutex. slows down execution a lot,
583 * but that's your problem, not mine.
584 */
585static mutex_t preadwritelock = X_MUTEX_INIT;
586
587static ssize_t 909static eio_ssize_t
588eio__pread (int fd, void *buf, size_t count, off_t offset) 910eio__pread (int fd, void *buf, size_t count, off_t offset)
589{ 911{
590 ssize_t res; 912 eio_ssize_t res;
591 off_t ooffset; 913 off_t ooffset;
592 914
593 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
594 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
595 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
… …
598 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
599 921
600 return res; 922 return res;
601} 923}
602 924
603static ssize_t 925static eio_ssize_t
604eio__pwrite (int fd, void *buf, size_t count, off_t offset) 926eio__pwrite (int fd, void *buf, size_t count, off_t offset)
605{ 927{
606 ssize_t res; 928 eio_ssize_t res;
607 off_t ooffset; 929 off_t ooffset;
608 930
609 X_LOCK (preadwritelock); 931 X_LOCK (preadwritelock);
610 ooffset = lseek (fd, 0, SEEK_CUR); 932 ooffset = lseek (fd, 0, SEEK_CUR);
611 lseek (fd, offset, SEEK_SET); 933 lseek (fd, offset, SEEK_SET);
612 res = write (fd, buf, count); 934 res = write (fd, buf, count);
613 lseek (fd, offset, SEEK_SET); 935 lseek (fd, ooffset, SEEK_SET);
614 X_UNLOCK (preadwritelock); 936 X_UNLOCK (preadwritelock);
615 937
616 return res; 938 return res;
617} 939}
618#endif 940#endif
619 941
620#ifndef HAVE_FUTIMES 942#ifndef HAVE_UTIMES
621 943
944# undef utimes
622# define utimes(path,times) eio__utimes (path, times) 945# define utimes(path,times) eio__utimes (path, times)
623# define futimes(fd,times) eio__futimes (fd, times)
624 946
625static int 947static int
626eio__utimes (const char *filename, const struct timeval times[2]) 948eio__utimes (const char *filename, const struct timeval times[2])
627{ 949{
628 if (times) 950 if (times)
… …
636 } 958 }
637 else 959 else
638 return utime (filename, 0); 960 return utime (filename, 0);
639} 961}
640 962
963#endif
964
965#ifndef HAVE_FUTIMES
966
967# undef futimes
968# define futimes(fd,times) eio__futimes (fd, times)
969
970static int
641static int eio__futimes (int fd, const struct timeval tv[2]) 971eio__futimes (int fd, const struct timeval tv[2])
642{ 972{
643 errno = ENOSYS; 973 errno = ENOSYS;
644 return -1; 974 return -1;
645} 975}
646 976
647#endif 977#endif
648 978
649#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
980# undef fdatasync
650# define fdatasync fsync 981# define fdatasync(fd) fsync (fd)
651#endif 982#endif
983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
1001
1002/* sync_file_range always needs emulation */
1003static int
1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1005{
1006#if HAVE_SYNC_FILE_RANGE
1007 int res;
1008
1009 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1010 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1011 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1012 {
1013 flags = 0
1014 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1015 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1016 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1017 }
1018
1019 res = sync_file_range (fd, offset, nbytes, flags);
1020
1021 if (!res || errno != ENOSYS)
1022 return res;
1023#endif
1024
1025 /* even though we could play tricks with the flags, it's better to always
1026 * call fdatasync, as that matches the expectation of its users best */
1027 return fdatasync (fd);
1028}
1029
1030static int
1031eio__fallocate (int fd, int mode, off_t offset, size_t len)
1032{
1033#if HAVE_FALLOCATE
1034 return fallocate (fd, mode, offset, len);
1035#else
1036 errno = ENOSYS;
1037 return -1;
1038#endif
1039}
652 1040
653#if !HAVE_READAHEAD 1041#if !HAVE_READAHEAD
1042# undef readahead
654# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1043# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655 1044
656static ssize_t 1045static eio_ssize_t
657eio__readahead (int fd, off_t offset, size_t count, worker *self) 1046eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658{ 1047{
659 size_t todo = count; 1048 size_t todo = count;
660 dBUF; 1049 dBUF;
661 1050
662 while (todo > 0) 1051 while (todo > 0)
… …
666 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
667 offset += len; 1056 offset += len;
668 todo -= len; 1057 todo -= len;
669 } 1058 }
670 1059
1060 FUBd;
1061
671 errno = 0; 1062 errno = 0;
672 return count; 1063 return count;
673} 1064}
674 1065
675#endif 1066#endif
676 1067
677#if !HAVE_READDIR_R
678# define readdir_r eio__readdir_r
679
680static mutex_t readdirlock = X_MUTEX_INIT;
681
682static int
683eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
684{
685 EIO_STRUCT_DIRENT *e;
686 int errorno;
687
688 X_LOCK (readdirlock);
689
690 e = readdir (dirp);
691 errorno = errno;
692
693 if (e)
694 {
695 *res = ent;
696 strcpy (ent->d_name, e->d_name);
697 }
698 else
699 *res = 0;
700
701 X_UNLOCK (readdirlock);
702
703 errno = errorno;
704 return e ? 0 : -1;
705}
706#endif
707
708/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
709static ssize_t 1069static eio_ssize_t
710eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
711{ 1071{
1072 eio_ssize_t written = 0;
712 ssize_t res; 1073 eio_ssize_t res;
713 1074
714 if (!count) 1075 if (!count)
715 return 0; 1076 return 0;
716 1077
1078 for (;;)
1079 {
1080#ifdef __APPLE__
1081# undef HAVE_SENDFILE /* broken, as everything on os x */
1082#endif
717#if HAVE_SENDFILE 1083#if HAVE_SENDFILE
718# if __linux 1084# if __linux
1085 off_t soffset = offset;
719 res = sendfile (ofd, ifd, &offset, count); 1086 res = sendfile (ofd, ifd, &soffset, count);
720 1087
721# elif __freebsd 1088# elif __FreeBSD__
722 /* 1089 /*
723 * Of course, the freebsd sendfile is a dire hack with no thoughts 1090 * Of course, the freebsd sendfile is a dire hack with no thoughts
724 * wasted on making it similar to other I/O functions. 1091 * wasted on making it similar to other I/O functions.
725 */ 1092 */
726 {
727 off_t sbytes; 1093 off_t sbytes;
728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1094 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729 1095
730 if (res < 0 && sbytes) 1096 #if 0 /* according to the manpage, this is correct, but broken behaviour */
731 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1097 /* freebsd' sendfile will return 0 on success */
1098 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1099 /* not on e.g. EIO or EPIPE - sounds broken */
1100 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
732 res = sbytes; 1101 res = sbytes;
733 } 1102 #endif
1103
1104 /* according to source inspection, this is correct, and useful behaviour */
1105 if (sbytes)
1106 res = sbytes;
1107
1108# elif defined (__APPLE__)
1109 off_t sbytes = count;
1110 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1111
1112 /* according to the manpage, sbytes is always valid */
1113 if (sbytes)
1114 res = sbytes;
734 1115
735# elif __hpux 1116# elif __hpux
736 res = sendfile (ofd, ifd, offset, count, 0, 0); 1117 res = sendfile (ofd, ifd, offset, count, 0, 0);
737 1118
738# elif __solaris 1119# elif __solaris
739 {
740 struct sendfilevec vec; 1120 struct sendfilevec vec;
741 size_t sbytes; 1121 size_t sbytes;
742 1122
743 vec.sfv_fd = ifd; 1123 vec.sfv_fd = ifd;
744 vec.sfv_flag = 0; 1124 vec.sfv_flag = 0;
745 vec.sfv_off = offset; 1125 vec.sfv_off = offset;
746 vec.sfv_len = count; 1126 vec.sfv_len = count;
747 1127
748 res = sendfilev (ofd, &vec, 1, &sbytes); 1128 res = sendfilev (ofd, &vec, 1, &sbytes);
749 1129
750 if (res < 0 && sbytes) 1130 if (res < 0 && sbytes)
751 res = sbytes; 1131 res = sbytes;
752 }
753 1132
754# endif 1133# endif
1134
1135#elif defined (_WIN32) && 0
1136 /* does not work, just for documentation of what would need to be done */
1137 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1138 /* libeio guarantees that the file offset does not change, and windows */
1139 /* has no way to get an independent handle to the same file description */
1140 HANDLE h = TO_SOCKET (ifd);
1141 SetFilePointer (h, offset, 0, FILE_BEGIN);
1142 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1143
755#else 1144#else
756 res = -1; 1145 res = -1;
757 errno = ENOSYS; 1146 errno = ENOSYS;
758#endif 1147#endif
759 1148
1149 /* we assume sendfile can copy at least 128mb in one go */
1150 if (res <= 128 * 1024 * 1024)
1151 {
1152 if (res > 0)
1153 written += res;
1154
1155 if (written)
1156 return written;
1157
1158 break;
1159 }
1160 else
1161 {
1162 /* if we requested more, then probably the kernel was lazy */
1163 written += res;
1164 offset += res;
1165 count -= res;
1166
1167 if (!count)
1168 return written;
1169 }
1170 }
1171
760 if (res < 0 1172 if (res < 0
761 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1173 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1174 /* BSDs */
1175#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1176 || errno == ENOTSUP
1177#endif
1178#ifdef EOPNOTSUPP /* windows */
1179 || errno == EOPNOTSUPP /* BSDs */
1180#endif
762#if __solaris 1181#if __solaris
763 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1182 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
764#endif 1183#endif
765 ) 1184 )
766 ) 1185 )
… …
770 1189
771 res = 0; 1190 res = 0;
772 1191
773 while (count) 1192 while (count)
774 { 1193 {
775 ssize_t cnt; 1194 eio_ssize_t cnt;
776 1195
777 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1196 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
778 1197
779 if (cnt <= 0) 1198 if (cnt <= 0)
780 { 1199 {
… …
792 1211
793 offset += cnt; 1212 offset += cnt;
794 res += cnt; 1213 res += cnt;
795 count -= cnt; 1214 count -= cnt;
796 } 1215 }
1216
1217 FUBd;
797 } 1218 }
798 1219
799 return res; 1220 return res;
1221}
1222
1223#ifdef PAGESIZE
1224# define eio_pagesize() PAGESIZE
1225#else
1226static intptr_t
1227eio_pagesize (void)
1228{
1229 static intptr_t page;
1230
1231 if (!page)
1232 page = sysconf (_SC_PAGESIZE);
1233
1234 return page;
1235}
1236#endif
1237
1238static void
1239eio_page_align (void **addr, size_t *length)
1240{
1241 intptr_t mask = eio_pagesize () - 1;
1242
1243 /* round down addr */
1244 intptr_t adj = mask & (intptr_t)*addr;
1245
1246 *addr = (void *)((intptr_t)*addr - adj);
1247 *length += adj;
1248
1249 /* round up length */
1250 *length = (*length + mask) & ~mask;
1251}
1252
1253#if !_POSIX_MEMLOCK
1254# define eio__mlockall(a) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__mlockall (int flags)
1259{
1260 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1261 extern int mallopt (int, int);
1262 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1263 #endif
1264
1265 if (EIO_MCL_CURRENT != MCL_CURRENT
1266 || EIO_MCL_FUTURE != MCL_FUTURE)
1267 {
1268 flags = 0
1269 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1270 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1271 }
1272
1273 return mlockall (flags);
1274}
1275#endif
1276
1277#if !_POSIX_MEMLOCK_RANGE
1278# define eio__mlock(a,b) EIO_ENOSYS ()
1279#else
1280
1281static int
1282eio__mlock (void *addr, size_t length)
1283{
1284 eio_page_align (&addr, &length);
1285
1286 return mlock (addr, length);
1287}
1288
1289#endif
1290
1291#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1292# define eio__msync(a,b,c) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__msync (void *mem, size_t len, int flags)
1297{
1298 eio_page_align (&mem, &len);
1299
1300 if (EIO_MS_ASYNC != MS_SYNC
1301 || EIO_MS_INVALIDATE != MS_INVALIDATE
1302 || EIO_MS_SYNC != MS_SYNC)
1303 {
1304 flags = 0
1305 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1306 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1307 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1308 }
1309
1310 return msync (mem, len, flags);
1311}
1312
1313#endif
1314
1315static int
1316eio__mtouch (eio_req *req)
1317{
1318 void *mem = req->ptr2;
1319 size_t len = req->size;
1320 int flags = req->int1;
1321
1322 eio_page_align (&mem, &len);
1323
1324 {
1325 intptr_t addr = (intptr_t)mem;
1326 intptr_t end = addr + len;
1327 intptr_t page = eio_pagesize ();
1328
1329 if (addr < end)
1330 if (flags & EIO_MT_MODIFY) /* modify */
1331 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1332 else
1333 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1334 }
1335
1336 return 0;
1337}
1338
1339/*****************************************************************************/
1340/* requests implemented outside eio_execute, because they are so large */
1341
1342static void
1343eio__realpath (eio_req *req, etp_worker *self)
1344{
1345 char *rel = req->ptr1;
1346 char *res;
1347 char *tmp1, *tmp2;
1348#if SYMLOOP_MAX > 32
1349 int symlinks = SYMLOOP_MAX;
1350#else
1351 int symlinks = 32;
1352#endif
1353
1354 req->result = -1;
1355
1356 errno = EINVAL;
1357 if (!rel)
1358 return;
1359
1360 errno = ENOENT;
1361 if (!*rel)
1362 return;
1363
1364 if (!req->ptr2)
1365 {
1366 X_LOCK (wrklock);
1367 req->flags |= EIO_FLAG_PTR2_FREE;
1368 X_UNLOCK (wrklock);
1369 req->ptr2 = malloc (PATH_MAX * 3);
1370
1371 errno = ENOMEM;
1372 if (!req->ptr2)
1373 return;
1374 }
1375
1376 res = req->ptr2;
1377 tmp1 = res + PATH_MAX;
1378 tmp2 = tmp1 + PATH_MAX;
1379
1380#if 0 /* disabled, the musl way to do things is just too racy */
1381#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 /* on linux we may be able to ask the kernel */
1383 {
1384 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1385
1386 if (fd >= 0)
1387 {
1388 sprintf (tmp1, "/proc/self/fd/%d", fd);
1389 req->result = readlink (tmp1, res, PATH_MAX);
1390 close (fd);
1391
1392 /* here we should probably stat the open file and the disk file, to make sure they still match */
1393
1394 if (req->result > 0)
1395 goto done;
1396 }
1397 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1398 return;
1399 }
1400#endif
1401#endif
1402
1403 if (*rel != '/')
1404 {
1405 if (!getcwd (res, PATH_MAX))
1406 return;
1407
1408 if (res [1]) /* only use if not / */
1409 res += strlen (res);
1410 }
1411
1412 while (*rel)
1413 {
1414 eio_ssize_t len, linklen;
1415 char *beg = rel;
1416
1417 while (*rel && *rel != '/')
1418 ++rel;
1419
1420 len = rel - beg;
1421
1422 if (!len) /* skip slashes */
1423 {
1424 ++rel;
1425 continue;
1426 }
1427
1428 if (beg [0] == '.')
1429 {
1430 if (len == 1)
1431 continue; /* . - nop */
1432
1433 if (beg [1] == '.' && len == 2)
1434 {
1435 /* .. - back up one component, if possible */
1436
1437 while (res != req->ptr2)
1438 if (*--res == '/')
1439 break;
1440
1441 continue;
1442 }
1443 }
1444
1445 errno = ENAMETOOLONG;
1446 if (res + 1 + len + 1 >= tmp1)
1447 return;
1448
1449 /* copy one component */
1450 *res = '/';
1451 memcpy (res + 1, beg, len);
1452
1453 /* zero-terminate, for readlink */
1454 res [len + 1] = 0;
1455
1456 /* now check if it's a symlink */
1457 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1458
1459 if (linklen < 0)
1460 {
1461 if (errno != EINVAL)
1462 return;
1463
1464 /* it's a normal directory. hopefully */
1465 res += len + 1;
1466 }
1467 else
1468 {
1469 /* yay, it was a symlink - build new path in tmp2 */
1470 int rellen = strlen (rel);
1471
1472 errno = ENAMETOOLONG;
1473 if (linklen + 1 + rellen >= PATH_MAX)
1474 return;
1475
1476 errno = ELOOP;
1477 if (!--symlinks)
1478 return;
1479
1480 if (*tmp1 == '/')
1481 res = req->ptr2; /* symlink resolves to an absolute path */
1482
1483 /* we need to be careful, as rel might point into tmp2 already */
1484 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1485 tmp2 [linklen] = '/';
1486 memcpy (tmp2, tmp1, linklen);
1487
1488 rel = tmp2;
1489 }
1490 }
1491
1492 /* special case for the lone root path */
1493 if (res == req->ptr2)
1494 *res++ = '/';
1495
1496 req->result = res - (char *)req->ptr2;
1497
1498done:
1499 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1500}
1501
1502static signed char
1503eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1504{
1505 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1506 : a->inode < b->inode ? -1
1507 : a->inode > b->inode ? 1
1508 : 0;
1509}
1510
1511#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1512
1513#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1515
1516static void
1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1518{
1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1520 unsigned char *bit = bits;
1521
1522 assert (CHAR_BIT == 8);
1523 assert (sizeof (eio_dirent) * 8 < 256);
1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1525 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1526
1527 if (size <= EIO_SORT_FAST)
1528 return;
1529
1530 /* first prepare an array of bits to test in our radix sort */
1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1532 /* inode_bits must contain all inodes ORed together */
1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1534 {
1535 eio_ino_t endianness;
1536 int i, j;
1537
1538 /* we store the byte offset of byte n into byte n of "endianness" */
1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1540 ((unsigned char *)&endianness)[i] = i;
1541
1542 *bit++ = 0;
1543
1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1545 {
1546 /* shifting off the byte offsets out of "endianness" */
1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1548 endianness >>= 8;
1549
1550 for (j = 0; j < 8; ++j)
1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1552 *bit++ = offs + j;
1553 }
1554
1555 for (j = 0; j < 8; ++j)
1556 if (score_bits & (1 << j))
1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1558 }
1559
1560 /* now actually do the sorting (a variant of MSD radix sort) */
1561 {
1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1565 int stk_idx = 0;
1566
1567 base_stk [stk_idx] = dents;
1568 end_stk [stk_idx] = dents + size;
1569 bit_stk [stk_idx] = bit - 1;
1570
1571 do
1572 {
1573 base = base_stk [stk_idx];
1574 end = end_stk [stk_idx];
1575 bit = bit_stk [stk_idx];
1576
1577 for (;;)
1578 {
1579 unsigned char O = *bit >> 3;
1580 unsigned char M = 1 << (*bit & 7);
1581
1582 eio_dirent *a = base;
1583 eio_dirent *b = end;
1584
1585 if (b - a < EIO_SORT_CUTOFF)
1586 break;
1587
1588 /* now bit-partition the array on the bit */
1589 /* this ugly asymmetric loop seems to perform much better than typical */
1590 /* partition algos found in the literature */
1591 do
1592 if (!(((unsigned char *)a)[O] & M))
1593 ++a;
1594 else if (!(((unsigned char *)--b)[O] & M))
1595 {
1596 eio_dirent tmp = *a; *a = *b; *b = tmp;
1597 ++a;
1598 }
1599 while (b > a);
1600
1601 /* next bit, or stop, if no bits left in this path */
1602 if (!*--bit)
1603 break;
1604
1605 base_stk [stk_idx] = a;
1606 end_stk [stk_idx] = end;
1607 bit_stk [stk_idx] = bit;
1608 ++stk_idx;
1609
1610 end = a;
1611 }
1612 }
1613 while (stk_idx--);
1614 }
1615}
1616
1617static void
1618eio_dent_insertion_sort (eio_dirent *dents, int size)
1619{
1620 /* first move the smallest element to the front, to act as a sentinel */
1621 {
1622 int i;
1623 eio_dirent *min = dents;
1624
1625 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1626 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1627 if (EIO_DENT_CMP (dents [i], <, *min))
1628 min = &dents [i];
1629
1630 /* swap elements 0 and j (minimum) */
1631 {
1632 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1633 }
1634 }
1635
1636 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1637 {
1638 eio_dirent *i, *j;
1639
1640 for (i = dents + 1; i < dents + size; ++i)
1641 {
1642 eio_dirent value = *i;
1643
1644 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1645 j [1] = j [0];
1646
1647 j [1] = value;
1648 }
1649 }
1650}
1651
1652static void
1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1654{
1655 if (size <= 1)
1656 return; /* our insertion sort relies on size > 0 */
1657
1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1659 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1660 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1661
1662 /* use an insertion sort at the end, or for small arrays, */
1663 /* as insertion sort is more efficient for small partitions */
1664 eio_dent_insertion_sort (dents, size);
800} 1665}
801 1666
802/* read a full directory */ 1667/* read a full directory */
803static void 1668static void
804eio__scandir (eio_req *req, worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
805{ 1670{
1671 char *name, *names;
1672 int namesalloc = 4096 - sizeof (void *) * 4;
1673 int namesoffs = 0;
1674 int flags = req->int1;
1675 eio_dirent *dents = 0;
1676 int dentalloc = 128;
1677 int dentoffs = 0;
1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
806 DIR *dirp; 1683 DIR *dirp;
807 union
808 {
809 EIO_STRUCT_DIRENT d;
810 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
811 } *u;
812 EIO_STRUCT_DIRENT *entp; 1684 EIO_STRUCT_DIRENT *entp;
813 char *name, *names; 1685#endif
814 int memlen = 4096;
815 int memofs = 0;
816 int res = 0;
817 1686
818 X_LOCK (etplock); 1687 req->result = -1;
1688
1689 if (!(flags & EIO_READDIR_DENTS))
1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1691
1692#ifdef _WIN32
1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 dirp = 0;
1712
1713 /* should steal _dosmaperr */
1714 switch (GetLastError ())
1715 {
1716 case ERROR_FILE_NOT_FOUND:
1717 req->result = 0;
1718 break;
1719
1720 case ERROR_INVALID_NAME:
1721 case ERROR_PATH_NOT_FOUND:
1722 case ERROR_NO_MORE_FILES:
1723 errno = ENOENT;
1724 break;
1725
1726 case ERROR_NOT_ENOUGH_MEMORY:
1727 errno = ENOMEM;
1728 break;
1729
1730 default:
1731 errno = EINVAL;
1732 break;
1733 }
1734 }
1735 }
1736#else
819 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
820 self->dbuf = u = malloc (sizeof (*u)); 1738#endif
1739
821 req->flags |= EIO_FLAG_PTR2_FREE; 1740 if (req->flags & EIO_FLAG_PTR1_FREE)
1741 free (req->ptr1);
1742
1743 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1744 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822 req->ptr2 = names = malloc (memlen); 1745 req->ptr2 = names = malloc (namesalloc);
823 X_UNLOCK (etplock);
824 1746
825 if (dirp && u && names) 1747 if (dirp && names && (!flags || dents))
826 for (;;) 1748 for (;;)
827 { 1749 {
1750 int done;
1751
1752#ifdef _WIN32
1753 done = !dirp;
1754#else
828 errno = 0; 1755 errno = 0;
829 readdir_r (dirp, &u->d, &entp); 1756 entp = readdir (dirp);
1757 done = !entp;
1758#endif
830 1759
831 if (!entp) 1760 if (done)
1761 {
1762#ifndef _WIN32
1763 int old_errno = errno;
1764 closedir (dirp);
1765 errno = old_errno;
1766
1767 if (errno)
1768 break;
1769#endif
1770
1771 /* sort etc. */
1772 req->int1 = flags;
1773 req->result = dentoffs;
1774
1775 if (flags & EIO_READDIR_STAT_ORDER)
1776 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1777 else if (flags & EIO_READDIR_DIRS_FIRST)
1778 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1779 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1780 else
1781 {
1782 /* in this case, all is known, and we just put dirs first and sort them */
1783 eio_dirent *oth = dents + dentoffs;
1784 eio_dirent *dir = dents;
1785
1786 /* now partition dirs to the front, and non-dirs to the back */
1787 /* by walking from both sides and swapping if necessary */
1788 while (oth > dir)
1789 {
1790 if (dir->type == EIO_DT_DIR)
1791 ++dir;
1792 else if ((--oth)->type == EIO_DT_DIR)
1793 {
1794 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1795
1796 ++dir;
1797 }
1798 }
1799
1800 /* now sort the dirs only (dirs all have the same score) */
1801 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1802 }
1803
832 break; 1804 break;
1805 }
833 1806
834 name = entp->d_name; 1807 /* now add the entry to our list(s) */
1808 name = D_NAME (entp);
835 1809
1810 /* skip . and .. entries */
836 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1811 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837 { 1812 {
838 int len = strlen (name) + 1; 1813 int len = D_NAMLEN (entp) + 1;
839 1814
840 res++; 1815 while (ecb_expect_false (namesoffs + len > namesalloc))
841
842 while (memofs + len > memlen)
843 { 1816 {
844 memlen *= 2; 1817 namesalloc *= 2;
845 X_LOCK (etplock);
846 req->ptr2 = names = realloc (names, memlen); 1818 req->ptr2 = names = realloc (names, namesalloc);
847 X_UNLOCK (etplock);
848 1819
849 if (!names) 1820 if (!names)
850 break; 1821 break;
851 } 1822 }
852 1823
853 memcpy (names + memofs, name, len); 1824 memcpy (names + namesoffs, name, len);
1825
1826 if (dents)
1827 {
1828 struct eio_dirent *ent;
1829
1830 if (ecb_expect_false (dentoffs == dentalloc))
1831 {
1832 dentalloc *= 2;
1833 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1834
1835 if (!dents)
1836 break;
1837 }
1838
1839 ent = dents + dentoffs;
1840
1841 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1842 ent->namelen = len - 1;
1843 ent->inode = D_INO (entp);
1844
1845 inode_bits |= ent->inode;
1846
1847 switch (D_TYPE (entp))
1848 {
1849 default:
1850 ent->type = EIO_DT_UNKNOWN;
1851 flags |= EIO_READDIR_FOUND_UNKNOWN;
1852 break;
1853
1854 #ifdef DT_FIFO
1855 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1856 #endif
1857 #ifdef DT_CHR
1858 case DT_CHR: ent->type = EIO_DT_CHR; break;
1859 #endif
1860 #ifdef DT_MPC
1861 case DT_MPC: ent->type = EIO_DT_MPC; break;
1862 #endif
1863 #ifdef DT_DIR
1864 case DT_DIR: ent->type = EIO_DT_DIR; break;
1865 #endif
1866 #ifdef DT_NAM
1867 case DT_NAM: ent->type = EIO_DT_NAM; break;
1868 #endif
1869 #ifdef DT_BLK
1870 case DT_BLK: ent->type = EIO_DT_BLK; break;
1871 #endif
1872 #ifdef DT_MPB
1873 case DT_MPB: ent->type = EIO_DT_MPB; break;
1874 #endif
1875 #ifdef DT_REG
1876 case DT_REG: ent->type = EIO_DT_REG; break;
1877 #endif
1878 #ifdef DT_NWK
1879 case DT_NWK: ent->type = EIO_DT_NWK; break;
1880 #endif
1881 #ifdef DT_CMP
1882 case DT_CMP: ent->type = EIO_DT_CMP; break;
1883 #endif
1884 #ifdef DT_LNK
1885 case DT_LNK: ent->type = EIO_DT_LNK; break;
1886 #endif
1887 #ifdef DT_SOCK
1888 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1889 #endif
1890 #ifdef DT_DOOR
1891 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1892 #endif
1893 #ifdef DT_WHT
1894 case DT_WHT: ent->type = EIO_DT_WHT; break;
1895 #endif
1896 }
1897
1898 ent->score = 7;
1899
1900 if (flags & EIO_READDIR_DIRS_FIRST)
1901 {
1902 if (ent->type == EIO_DT_UNKNOWN)
1903 {
1904 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1905 ent->score = 1;
1906 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1907 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1908 }
1909 else if (ent->type == EIO_DT_DIR)
1910 ent->score = 0;
1911 }
1912 }
1913
854 memofs += len; 1914 namesoffs += len;
1915 ++dentoffs;
855 } 1916 }
1917
1918 if (EIO_CANCELLED (req))
1919 {
1920 errno = ECANCELED;
1921 break;
1922 }
1923
1924#ifdef _WIN32
1925 if (!FindNextFile (dirp, &entp))
1926 {
1927 FindClose (dirp);
1928 dirp = 0;
1929 }
1930#endif
856 } 1931 }
857
858 if (errno)
859 res = -1;
860
861 req->result = res;
862} 1932}
863 1933
864/*****************************************************************************/ 1934/*****************************************************************************/
865 1935
866#define ALLOC(len) \ 1936#define ALLOC(len) \
867 if (!req->ptr2) \ 1937 if (!req->ptr2) \
868 { \ 1938 { \
869 X_LOCK (etplock); \ 1939 X_LOCK (wrklock); \
870 req->flags |= EIO_FLAG_PTR2_FREE; \ 1940 req->flags |= EIO_FLAG_PTR2_FREE; \
871 X_UNLOCK (etplock); \ 1941 X_UNLOCK (wrklock); \
872 req->ptr2 = malloc (len); \ 1942 req->ptr2 = malloc (len); \
873 if (!req->ptr2) \ 1943 if (!req->ptr2) \
874 { \ 1944 { \
875 errno = ENOMEM; \ 1945 errno = ENOMEM; \
876 req->result = -1; \ 1946 req->result = -1; \
877 break; \ 1947 break; \
878 } \ 1948 } \
879 } 1949 }
880 1950
881X_THREAD_PROC (eio_proc) 1951X_THREAD_PROC (etp_proc)
882{ 1952{
883 eio_req *req; 1953 ETP_REQ *req;
884 struct timespec ts; 1954 struct timespec ts;
885 worker *self = (worker *)thr_arg; 1955 etp_worker *self = (etp_worker *)thr_arg;
886 1956
1957#if HAVE_PRCTL_SET_NAME
1958 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1959#endif
1960
887 /* try to distribute timeouts somewhat randomly */ 1961 /* try to distribute timeouts somewhat evenly */
888 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1962 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889 1963
890 for (;;) 1964 for (;;)
891 { 1965 {
892 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1966 ts.tv_sec = 0;
893 1967
894 X_LOCK (reqlock); 1968 X_LOCK (reqlock);
895 1969
896 for (;;) 1970 for (;;)
897 { 1971 {
898 self->req = req = reqq_shift (&req_queue); 1972 self->req = req = reqq_shift (&req_queue);
899 1973
900 if (req) 1974 if (req)
901 break; 1975 break;
902 1976
1977 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1978 {
1979 X_UNLOCK (reqlock);
1980 X_LOCK (wrklock);
1981 --started;
1982 X_UNLOCK (wrklock);
1983 goto quit;
1984 }
1985
903 ++idle; 1986 ++idle;
904 1987
905 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1988 if (idle <= max_idle)
1989 /* we are allowed to idle, so do so without any timeout */
1990 X_COND_WAIT (reqwait, reqlock);
1991 else
906 { 1992 {
907 if (idle > max_idle) 1993 /* initialise timeout once */
908 { 1994 if (!ts.tv_sec)
909 --idle; 1995 ts.tv_sec = time (0) + idle_timeout;
910 X_UNLOCK (reqlock);
911 X_LOCK (etplock);
912 --started;
913 X_UNLOCK (etplock);
914 goto quit;
915 }
916 1996
917 /* we are allowed to idle, so do so without any timeout */
918 X_COND_WAIT (reqwait, reqlock); 1997 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
919 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1998 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
920 } 1999 }
921 2000
922 --idle; 2001 --idle;
923 } 2002 }
924 2003
925 --nready; 2004 --nready;
926 2005
927 X_UNLOCK (reqlock); 2006 X_UNLOCK (reqlock);
928 2007
929 errno = 0; /* strictly unnecessary */
930
931 if (!EIO_CANCELLED (req))
932 switch (req->type) 2008 if (req->type < 0)
933 {
934 case EIO_READ: ALLOC (req->size);
935 req->result = req->offs >= 0
936 ? pread (req->int1, req->ptr2, req->size, req->offs)
937 : read (req->int1, req->ptr2, req->size); break;
938 case EIO_WRITE: req->result = req->offs >= 0
939 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
940 : write (req->int1, req->ptr2, req->size); break;
941
942 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
943 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
944
945 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
946 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
947 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
948 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
949 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
950 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
951
952 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
953 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
954 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
955 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
956 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
957 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
958
959 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
960 case EIO_CLOSE: req->result = close (req->int1); break;
961 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
962 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
963 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
964 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
965 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
966 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
967 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
968 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
969
970 case EIO_READLINK: ALLOC (NAME_MAX);
971 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
972
973 case EIO_SYNC: req->result = 0; sync (); break;
974 case EIO_FSYNC: req->result = fsync (req->int1); break;
975 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
976
977 case EIO_READDIR: eio__scandir (req, self); break;
978
979 case EIO_BUSY:
980#ifdef _WIN32
981 Sleep (req->nv1 * 1000.);
982#else
983 {
984 struct timeval tv;
985
986 tv.tv_sec = req->nv1;
987 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
988
989 req->result = select (0, 0, 0, 0, &tv);
990 }
991#endif
992 break;
993
994 case EIO_UTIME:
995 case EIO_FUTIME:
996 {
997 struct timeval tv[2];
998 struct timeval *times;
999
1000 if (req->nv1 != -1. || req->nv2 != -1.)
1001 {
1002 tv[0].tv_sec = req->nv1;
1003 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1004 tv[1].tv_sec = req->nv2;
1005 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1006
1007 times = tv;
1008 }
1009 else
1010 times = 0;
1011
1012
1013 req->result = req->type == EIO_FUTIME
1014 ? futimes (req->int1, times)
1015 : utimes (req->ptr1, times);
1016 }
1017
1018 case EIO_GROUP:
1019 case EIO_NOP:
1020 req->result = 0;
1021 break;
1022
1023 case EIO_QUIT:
1024 goto quit; 2009 goto quit;
1025 2010
1026 default: 2011 ETP_EXECUTE (self, req);
1027 req->result = -1;
1028 break;
1029 }
1030
1031 req->errorno = errno;
1032 2012
1033 X_LOCK (reslock); 2013 X_LOCK (reslock);
1034 2014
1035 ++npending; 2015 ++npending;
1036 2016
1037 if (!reqq_push (&res_queue, req) && etp.want_poll_cb) 2017 if (!reqq_push (&res_queue, req) && want_poll_cb)
1038 etp.want_poll_cb (); 2018 want_poll_cb ();
1039 2019
1040 self->req = 0; 2020 self->req = 0;
1041 worker_clear (self); 2021 etp_worker_clear (self);
1042 2022
1043 X_UNLOCK (reslock); 2023 X_UNLOCK (reslock);
1044 } 2024 }
1045 2025
1046quit: 2026quit:
1047 X_LOCK (etplock); 2027 X_LOCK (wrklock);
1048 worker_free (self); 2028 etp_worker_free (self);
1049 X_UNLOCK (etplock); 2029 X_UNLOCK (wrklock);
1050 2030
1051 return 0; 2031 return 0;
1052} 2032}
1053 2033
1054/*****************************************************************************/ 2034/*****************************************************************************/
1055 2035
2036int ecb_cold
1056int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2037eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057{ 2038{
2039#if !HAVE_PREADWRITE
2040 X_MUTEX_CREATE (preadwritelock);
2041#endif
2042
1058 etp_init (&etp, want_poll, done_poll); 2043 return etp_init (want_poll, done_poll);
1059} 2044}
1060 2045
2046ecb_inline void
1061static void eio_api_destroy (eio_req *req) 2047eio_api_destroy (eio_req *req)
1062{ 2048{
1063 free (req); 2049 free (req);
1064} 2050}
1065 2051
1066#define REQ(rtype) \ 2052#define REQ(rtype) \
… …
1085 { \ 2071 { \
1086 eio_api_destroy (req); \ 2072 eio_api_destroy (req); \
1087 return 0; \ 2073 return 0; \
1088 } 2074 }
1089 2075
2076static void
2077eio_execute (etp_worker *self, eio_req *req)
2078{
2079 if (ecb_expect_false (EIO_CANCELLED (req)))
2080 {
2081 req->result = -1;
2082 req->errorno = ECANCELED;
2083 return;
2084 }
2085
2086 switch (req->type)
2087 {
2088 case EIO_READ: ALLOC (req->size);
2089 req->result = req->offs >= 0
2090 ? pread (req->int1, req->ptr2, req->size, req->offs)
2091 : read (req->int1, req->ptr2, req->size); break;
2092 case EIO_WRITE: req->result = req->offs >= 0
2093 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2094 : write (req->int1, req->ptr2, req->size); break;
2095
2096 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2097 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2098
2099 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2100 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2101 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2102 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2103 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2104 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2105
2106 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2107 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2108 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2109 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2110
2111 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2112 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2113 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2114 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2115 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2116 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2117
2118 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2119 case EIO_CLOSE: req->result = close (req->int1); break;
2120 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2121 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2122 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2123 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2124 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2125 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2126 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2127 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2128
2129 case EIO_REALPATH: eio__realpath (req, self); break;
2130
2131 case EIO_READLINK: ALLOC (PATH_MAX);
2132 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2133
2134 case EIO_SYNC: req->result = 0; sync (); break;
2135 case EIO_FSYNC: req->result = fsync (req->int1); break;
2136 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2137 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2138 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2139 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2140 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2141 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2142 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2143 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2144
2145 case EIO_READDIR: eio__scandir (req, self); break;
2146
2147 case EIO_BUSY:
2148#ifdef _WIN32
2149 Sleep (req->nv1 * 1e3);
2150#else
2151 {
2152 struct timeval tv;
2153
2154 tv.tv_sec = req->nv1;
2155 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2156
2157 req->result = select (0, 0, 0, 0, &tv);
2158 }
2159#endif
2160 break;
2161
2162 case EIO_UTIME:
2163 case EIO_FUTIME:
2164 {
2165 struct timeval tv[2];
2166 struct timeval *times;
2167
2168 if (req->nv1 != -1. || req->nv2 != -1.)
2169 {
2170 tv[0].tv_sec = req->nv1;
2171 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2172 tv[1].tv_sec = req->nv2;
2173 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2174
2175 times = tv;
2176 }
2177 else
2178 times = 0;
2179
2180 req->result = req->type == EIO_FUTIME
2181 ? futimes (req->int1, times)
2182 : utimes (req->ptr1, times);
2183 }
2184 break;
2185
2186 case EIO_GROUP:
2187 abort (); /* handled in eio_request */
2188
2189 case EIO_NOP:
2190 req->result = 0;
2191 break;
2192
2193 case EIO_CUSTOM:
2194 req->feed (req);
2195 break;
2196
2197 default:
2198 errno = ENOSYS;
2199 req->result = -1;
2200 break;
2201 }
2202
2203 req->errorno = errno;
2204}
2205
1090#ifndef EIO_NO_WRAPPERS 2206#ifndef EIO_NO_WRAPPERS
1091 2207
1092eio_req *eio_nop (int pri, eio_cb cb, void *data) 2208eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093{ 2209{
1094 REQ (EIO_NOP); SEND; 2210 REQ (EIO_NOP); SEND;
… …
1107eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2223eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108{ 2224{
1109 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2225 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110} 2226}
1111 2227
2228eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2231}
2232
1112eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2233eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113{ 2234{
1114 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2235 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115} 2236}
1116 2237
2238eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2246}
2247
2248eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2249{
2250 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2251}
2252
2253eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2254{
2255 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2256}
2257
2258eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2259{
2260 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2261}
2262
2263eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2264{
2265 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2266}
2267
1117eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2268eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1118{ 2269{
1119 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2270 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1120} 2271}
1121 2272
… …
1137eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2288eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138{ 2289{
1139 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2290 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1140} 2291}
1141 2292
2293eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2294{
2295 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2296}
2297
1142eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2298eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143{ 2299{
1144 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2300 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1145} 2301}
1146 2302
… …
1152eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2308eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2309{
1154 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2310 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1155} 2311}
1156 2312
1157eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2313eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1158{ 2314{
1159 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2315 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1160} 2316}
1161 2317
1162eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2318eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1182eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2338eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1183{ 2339{
1184 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2340 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1185} 2341}
1186 2342
1187eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2343eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1188{ 2344{
1189 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2345 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1190} 2346}
1191 2347
1192eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2348eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1208eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2364eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1209{ 2365{
1210 return eio__1path (EIO_READLINK, path, pri, cb, data); 2366 return eio__1path (EIO_READLINK, path, pri, cb, data);
1211} 2367}
1212 2368
2369eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2370{
2371 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2372}
2373
1213eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2374eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1214{ 2375{
1215 return eio__1path (EIO_STAT, path, pri, cb, data); 2376 return eio__1path (EIO_STAT, path, pri, cb, data);
1216} 2377}
1217 2378
1218eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2379eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219{ 2380{
1220 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2381 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221} 2382}
1222 2383
2384eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2385{
2386 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2387}
2388
1223eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2389eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224{ 2390{
1225 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2391 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226} 2392}
1227 2393
1228eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2394eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229{ 2395{
1230 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2396 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231} 2397}
1232 2398
1233eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2399eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234{ 2400{
1235 return eio__1path (EIO_READDIR, path, pri, cb, data); 2401 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236} 2402}
1237 2403
1238eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2404eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239{ 2405{
1240 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2406 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1241} 2407}
1242 2408
1243static eio_req * 2409static eio_req *
1244eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2410eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245{ 2411{
… …
1269eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2435eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270{ 2436{
1271 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2437 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272} 2438}
1273 2439
2440eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2441{
2442 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2443}
2444
1274#endif 2445#endif
1275 2446
1276eio_req *eio_grp (eio_cb cb, void *data) 2447eio_req *eio_grp (eio_cb cb, void *data)
1277{ 2448{
1278 const int pri = EIO_PRI_MAX; 2449 const int pri = EIO_PRI_MAX;
… …
1285#undef SEND 2456#undef SEND
1286 2457
1287/*****************************************************************************/ 2458/*****************************************************************************/
1288/* grp functions */ 2459/* grp functions */
1289 2460
2461void
1290void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2462eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1291{ 2463{
1292 grp->int2 = limit; 2464 grp->int2 = limit;
1293 grp->feed = feed; 2465 grp->feed = feed;
1294 2466
1295 grp_try_feed (grp); 2467 grp_try_feed (grp);
1296} 2468}
1297 2469
2470void
1298void eio_grp_limit (eio_req *grp, int limit) 2471eio_grp_limit (eio_req *grp, int limit)
1299{ 2472{
1300 grp->int2 = limit; 2473 grp->int2 = limit;
1301 2474
1302 grp_try_feed (grp); 2475 grp_try_feed (grp);
1303} 2476}
1304 2477
2478void
1305void eio_grp_add (eio_req *grp, eio_req *req) 2479eio_grp_add (eio_req *grp, eio_req *req)
1306{ 2480{
1307 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2481 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2482
2483 grp->flags |= EIO_FLAG_GROUPADD;
1308 2484
1309 ++grp->size; 2485 ++grp->size;
1310 req->grp = grp; 2486 req->grp = grp;
1311 2487
1312 req->grp_prev = 0; 2488 req->grp_prev = 0;
… …
1319} 2495}
1320 2496
1321/*****************************************************************************/ 2497/*****************************************************************************/
1322/* misc garbage */ 2498/* misc garbage */
1323 2499
2500eio_ssize_t
1324ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2501eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325{ 2502{
1326 worker wrk;
1327
1328 wrk.dbuf = 0;
1329
1330 eio__sendfile (ofd, ifd, offset, count, &wrk); 2503 return eio__sendfile (ofd, ifd, offset, count);
1331
1332 if (wrk.dbuf)
1333 free (wrk.dbuf);
1334} 2504}
1335 2505

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines