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 1.99 by root, Tue Jul 26 11:07:08 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 #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 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu)
136
137 static int
138 eio__rename (const char *old, const char *neu)
139 {
140 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
141 return 0;
142
143 /* should steal _dosmaperr */
144 switch (GetLastError ())
145 {
146 case ERROR_FILE_NOT_FOUND:
147 case ERROR_PATH_NOT_FOUND:
148 case ERROR_INVALID_DRIVE:
149 case ERROR_NO_MORE_FILES:
150 case ERROR_BAD_NETPATH:
151 case ERROR_BAD_NET_NAME:
152 case ERROR_BAD_PATHNAME:
153 case ERROR_FILENAME_EXCED_RANGE:
154 errno = ENOENT;
155 break;
156
157 default:
158 errno = EACCES;
159 break;
160 }
161
162 return -1;
163 }
164
165 /* we could even stat and see if it exists */
166 static int
167 symlink (const char *old, const char *neu)
168 {
169 #if WINVER >= 0x0600
170 if (CreateSymbolicLink (neu, old, 1))
171 return 0;
172
173 if (CreateSymbolicLink (neu, old, 0))
174 return 0;
175 #endif
176
177 return EIO_ERRNO (ENOENT, -1);
178 }
179
180 /* POSIX API only */
181 #define CreateHardLink(neu,old,flags) 0
182 #define CreateSymbolicLink(neu,old,flags) 0
183
184 struct statvfs
185 {
186 int dummy;
187 };
188
189 #define DT_DIR EIO_DT_DIR
190 #define DT_REG EIO_DT_REG
191 #define D_NAME(entp) entp.cFileName
192 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 193
31#else 194#else
32 195
33# include "config.h"
34# include <sys/time.h> 196 #include <sys/time.h>
35# include <sys/select.h> 197 #include <sys/select.h>
198 #include <sys/statvfs.h>
36# include <unistd.h> 199 #include <unistd.h>
200 #include <signal.h>
201 #include <dirent.h>
202
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h>
205 #endif
206
207 #define D_NAME(entp) entp->d_name
208
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino
216 #endif
217
218 #ifdef _D_EXACT_NAMLEN
219 #undef D_NAMLEN
220 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
221 #endif
222
223 #ifdef _DIRENT_HAVE_D_TYPE
224 #define D_TYPE(de) (de)->d_type
225 #endif
226
227 #ifndef EIO_STRUCT_DIRENT
228 #define EIO_STRUCT_DIRENT struct dirent
229 #endif
230
231#endif
232
233#if HAVE_UTIMES
37# include <utime.h> 234# 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 235#endif
44 236
237#if HAVE_SYS_SYSCALL_H
238# include <sys/syscall.h>
239#endif
240
241#if HAVE_SYS_PRCTL_H
242# include <sys/prctl.h>
45#endif 243#endif
46 244
47#if HAVE_SENDFILE 245#if HAVE_SENDFILE
48# if __linux 246# if __linux
49# include <sys/sendfile.h> 247# include <sys/sendfile.h>
50# elif __freebsd 248# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 249# include <sys/socket.h>
52# include <sys/uio.h> 250# include <sys/uio.h>
53# elif __hpux 251# elif __hpux
54# include <sys/socket.h> 252# include <sys/socket.h>
55# elif __solaris /* not yet */ 253# elif __solaris
56# include <sys/sendfile.h> 254# include <sys/sendfile.h>
57# else 255# else
58# error sendfile support requested but not available 256# error sendfile support requested but not available
59# endif 257# endif
60#endif 258#endif
61 259
62/* number of seconds after which an idle threads exit */ 260#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 261# define D_TYPE(de) 0
262#endif
263#ifndef D_INO
264# define D_INO(de) 0
265#endif
266#ifndef D_NAMLEN
267# define D_NAMLEN(entp) strlen (D_NAME (entp))
268#endif
64 269
65/* used for struct dirent, AIX doesn't provide it */ 270/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 271#ifndef NAME_MAX
67# define NAME_MAX 4096 272# define NAME_MAX 4096
68#endif 273#endif
69 274
275/* used for readlink etc. */
276#ifndef PATH_MAX
277# define PATH_MAX 4096
278#endif
279
70/* buffer size for various temporary buffers */ 280/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 281#define EIO_BUFSIZE 65536
72 282
73#define dBUF \ 283#define dBUF \
74 char *eio_buf; \
75 X_LOCK (etplock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 284 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (etplock); \
78 errno = ENOMEM; \ 285 errno = ENOMEM; \
79 if (!eio_buf) \ 286 if (!eio_buf) \
80 return -1; 287 return -1
288
289#define FUBd \
290 free (eio_buf)
81 291
82#define EIO_TICKS ((1000000 + 1023) >> 10) 292#define EIO_TICKS ((1000000 + 1023) >> 10)
83 293
294#define ETP_PRI_MIN EIO_PRI_MIN
295#define ETP_PRI_MAX EIO_PRI_MAX
296
297struct etp_worker;
298
299#define ETP_REQ eio_req
300#define ETP_DESTROY(req) eio_destroy (req)
301static int eio_finish (eio_req *req);
302#define ETP_FINISH(req) eio_finish (req)
303static void eio_execute (struct etp_worker *self, eio_req *req);
304#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
305
84/*****************************************************************************/ 306/*****************************************************************************/
85 307
308#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
309
86/* calculcate time difference in ~1/EIO_TICKS of a second */ 310/* calculate time difference in ~1/EIO_TICKS of a second */
311ecb_inline int
87static int tvdiff (struct timeval *tv1, struct timeval *tv2) 312tvdiff (struct timeval *tv1, struct timeval *tv2)
88{ 313{
89 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 314 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 315 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
91} 316}
92 317
93static unsigned int started, idle, wanted = 4; 318static unsigned int started, idle, wanted = 4;
94 319
95typedef struct etp_pool
96{
97 void (*want_poll_cb) (void); 320static void (*want_poll_cb) (void);
98 void (*done_poll_cb) (void); 321static void (*done_poll_cb) (void);
99 322
100 unsigned int max_poll_time; 323static unsigned int max_poll_time; /* reslock */
101 unsigned int max_poll_reqs; 324static unsigned int max_poll_reqs; /* reslock */
102} etp_pool;
103 325
104static volatile unsigned int nreqs, nready, npending; 326static volatile unsigned int nreqs; /* reqlock */
105static volatile unsigned int max_idle = 4; 327static volatile unsigned int nready; /* reqlock */
328static volatile unsigned int npending; /* reqlock */
329static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
330static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
106 331
107static mutex_t etplock = X_MUTEX_INIT; 332static xmutex_t wrklock;
108static mutex_t reslock = X_MUTEX_INIT; 333static xmutex_t reslock;
109static mutex_t reqlock = X_MUTEX_INIT; 334static xmutex_t reqlock;
110static cond_t reqwait = X_COND_INIT; 335static xcond_t reqwait;
111 336
337#if !HAVE_PREADWRITE
338/*
339 * make our pread/pwrite emulation safe against themselves, but not against
340 * normal read/write by using a mutex. slows down execution a lot,
341 * but that's your problem, not mine.
342 */
343static xmutex_t preadwritelock;
344#endif
345
112typedef struct worker 346typedef struct etp_worker
113{ 347{
114 /* locked by etplock */ 348 /* locked by wrklock */
115 struct worker *prev, *next; 349 struct etp_worker *prev, *next;
116 350
117 thread_t tid; 351 xthread_t tid;
118 352
119 /* locked by reslock, reqlock or etplock */ 353 /* locked by reslock, reqlock or wrklock */
120 eio_req *req; /* currently processed request */ 354 ETP_REQ *req; /* currently processed request */
121 void *dbuf; 355
122 DIR *dirp; 356#ifdef ETP_WORKER_COMMON
357 ETP_WORKER_COMMON
358#endif
123} worker; 359} etp_worker;
124 360
125static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 361static etp_worker wrk_first; /* NOT etp */
362
363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126 365
127/* worker threads management */ 366/* worker threads management */
128 367
368static void ecb_cold
129static void worker_clear (worker *wrk) 369etp_worker_clear (etp_worker *wrk)
130{ 370{
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} 371}
143 372
144static void worker_free (worker *wrk) 373static void ecb_cold
374etp_worker_free (etp_worker *wrk)
145{ 375{
146 wrk->next->prev = wrk->prev; 376 wrk->next->prev = wrk->prev;
147 wrk->prev->next = wrk->next; 377 wrk->prev->next = wrk->next;
148 378
149 free (wrk); 379 free (wrk);
150} 380}
151 381
152unsigned int eio_nreqs (void) 382static unsigned int
383etp_nreqs (void)
153{ 384{
385 int retval;
386 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
387 retval = nreqs;
388 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 389 return retval;
155} 390}
156 391
157unsigned int eio_nready (void) 392static unsigned int
393etp_nready (void)
158{ 394{
159 unsigned int retval; 395 unsigned int retval;
160 396
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 397 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 398 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 399 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 400
165 return retval; 401 return retval;
166} 402}
167 403
168unsigned int eio_npending (void) 404static unsigned int
405etp_npending (void)
169{ 406{
170 unsigned int retval; 407 unsigned int retval;
171 408
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 409 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 410 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 411 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 412
176 return retval; 413 return retval;
177} 414}
178 415
179unsigned int eio_nthreads (void) 416static unsigned int
417etp_nthreads (void)
180{ 418{
181 unsigned int retval; 419 unsigned int retval;
182 420
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 421 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 422 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 429 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 430 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 431 * per shift, the most expensive operation.
194 */ 432 */
195typedef struct { 433typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 434 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 435 int size;
198} reqq; 436} etp_reqq;
199 437
200static reqq req_queue; 438static etp_reqq req_queue;
201static reqq res_queue; 439static etp_reqq res_queue;
202 440
203static int reqq_push (reqq *q, eio_req *req) 441static void ecb_noinline ecb_cold
442reqq_init (etp_reqq *q)
443{
444 int pri;
445
446 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
447 q->qs[pri] = q->qe[pri] = 0;
448
449 q->size = 0;
450}
451
452static int ecb_noinline
453reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 454{
205 int pri = req->pri; 455 int pri = req->pri;
206 req->next = 0; 456 req->next = 0;
207 457
208 if (q->qe[pri]) 458 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 464 q->qe[pri] = q->qs[pri] = req;
215 465
216 return q->size++; 466 return q->size++;
217} 467}
218 468
219static eio_req *reqq_shift (reqq *q) 469static ETP_REQ * ecb_noinline
470reqq_shift (etp_reqq *q)
220{ 471{
221 int pri; 472 int pri;
222 473
223 if (!q->size) 474 if (!q->size)
224 return 0; 475 return 0;
225 476
226 --q->size; 477 --q->size;
227 478
228 for (pri = EIO_NUM_PRI; pri--; ) 479 for (pri = ETP_NUM_PRI; pri--; )
229 { 480 {
230 eio_req *req = q->qs[pri]; 481 eio_req *req = q->qs[pri];
231 482
232 if (req) 483 if (req)
233 { 484 {
… …
239 } 490 }
240 491
241 abort (); 492 abort ();
242} 493}
243 494
244static void etp_atfork_prepare (void) 495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 497{
246 X_LOCK (etplock); 498 X_MUTEX_CREATE (wrklock);
247 X_LOCK (reqlock); 499 X_MUTEX_CREATE (reslock);
248 X_LOCK (reslock); 500 X_MUTEX_CREATE (reqlock);
249#if !HAVE_PREADWRITE 501 X_COND_CREATE (reqwait);
250 X_LOCK (preadwritelock);
251#endif
252#if !HAVE_READDIR_R
253 X_LOCK (readdirlock);
254#endif
255}
256 502
257static void etp_atfork_parent (void) 503 reqq_init (&req_queue);
258{ 504 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}
269 505
270static void etp_atfork_child (void) 506 wrk_first.next =
271{ 507 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 }
290 508
291 started = 0; 509 started = 0;
292 idle = 0; 510 idle = 0;
293 nreqs = 0; 511 nreqs = 0;
294 nready = 0; 512 nready = 0;
295 npending = 0; 513 npending = 0;
296 514
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; 515 want_poll_cb = want_poll;
316 etp->done_poll_cb = done_poll; 516 done_poll_cb = done_poll;
317}
318 517
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; 518 return 0;
354} 519}
355 520
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); 521X_THREAD_PROC (etp_proc);
407 522
408static void start_thread (void) 523static void ecb_cold
524etp_start_thread (void)
409{ 525{
410 worker *wrk = calloc (1, sizeof (worker)); 526 etp_worker *wrk = calloc (1, sizeof (etp_worker));
411 527
412 /*TODO*/ 528 /*TODO*/
413 assert (("unable to allocate worker thread data", wrk)); 529 assert (("unable to allocate worker thread data", wrk));
414 530
415 X_LOCK (etplock); 531 X_LOCK (wrklock);
416 532
417 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 533 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418 { 534 {
419 wrk->prev = &wrk_first; 535 wrk->prev = &wrk_first;
420 wrk->next = wrk_first.next; 536 wrk->next = wrk_first.next;
421 wrk_first.next->prev = wrk; 537 wrk_first.next->prev = wrk;
422 wrk_first.next = wrk; 538 wrk_first.next = wrk;
423 ++started; 539 ++started;
424 } 540 }
425 else 541 else
426 free (wrk); 542 free (wrk);
427 543
428 X_UNLOCK (etplock); 544 X_UNLOCK (wrklock);
429} 545}
430 546
547static void
431static void maybe_start_thread (void) 548etp_maybe_start_thread (void)
432{ 549{
433 if (eio_nthreads () >= wanted) 550 if (ecb_expect_true (etp_nthreads () >= wanted))
434 return; 551 return;
435 552
436 /* todo: maybe use idle here, but might be less exact */ 553 /* todo: maybe use idle here, but might be less exact */
437 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 554 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438 return; 555 return;
439 556
440 start_thread (); 557 etp_start_thread ();
441} 558}
442 559
443void eio_submit (eio_req *req) 560static void ecb_cold
444{ 561etp_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{ 562{
463 eio_req *req = calloc (1, sizeof (eio_req)); 563 eio_req *req = calloc (1, sizeof (eio_req));
464 564
465 req->type = EIO_QUIT; 565 req->type = -1;
466 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 566 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467 567
468 X_LOCK (reqlock); 568 X_LOCK (reqlock);
469 reqq_push (&req_queue, req); 569 reqq_push (&req_queue, req);
470 X_COND_SIGNAL (reqwait); 570 X_COND_SIGNAL (reqwait);
471 X_UNLOCK (reqlock); 571 X_UNLOCK (reqlock);
472 572
473 X_LOCK (etplock); 573 X_LOCK (wrklock);
474 --started; 574 --started;
475 X_UNLOCK (etplock); 575 X_UNLOCK (wrklock);
476} 576}
477 577
478void eio_set_max_poll_time (double nseconds) 578static 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) 579etp_poll (void)
515{ 580{
516 int maxreqs = etp.max_poll_reqs; 581 unsigned int maxreqs;
582 unsigned int maxtime;
517 struct timeval tv_start, tv_now; 583 struct timeval tv_start, tv_now;
518 eio_req *req;
519 584
520 if (etp.max_poll_time) 585 X_LOCK (reslock);
586 maxreqs = max_poll_reqs;
587 maxtime = max_poll_time;
588 X_UNLOCK (reslock);
589
590 if (maxtime)
521 gettimeofday (&tv_start, 0); 591 gettimeofday (&tv_start, 0);
522 592
523 for (;;) 593 for (;;)
524 { 594 {
595 ETP_REQ *req;
596
525 maybe_start_thread (); 597 etp_maybe_start_thread ();
526 598
527 X_LOCK (reslock); 599 X_LOCK (reslock);
528 req = reqq_shift (&res_queue); 600 req = reqq_shift (&res_queue);
529 601
530 if (req) 602 if (req)
531 { 603 {
532 --npending; 604 --npending;
533 605
534 if (!res_queue.size && etp.done_poll_cb) 606 if (!res_queue.size && done_poll_cb)
535 etp.done_poll_cb (); 607 done_poll_cb ();
536 } 608 }
537 609
538 X_UNLOCK (reslock); 610 X_UNLOCK (reslock);
539 611
540 if (!req) 612 if (!req)
541 return 0; 613 return 0;
542 614
615 X_LOCK (reqlock);
543 --nreqs; 616 --nreqs;
617 X_UNLOCK (reqlock);
544 618
545 if (req->type == EIO_GROUP && req->size) 619 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
546 { 620 {
547 req->int1 = 1; /* mark request as delayed */ 621 req->int1 = 1; /* mark request as delayed */
548 continue; 622 continue;
549 } 623 }
550 else 624 else
551 { 625 {
552 int res = eio_finish (req); 626 int res = ETP_FINISH (req);
553 if (res) 627 if (ecb_expect_false (res))
554 return res; 628 return res;
555 } 629 }
556 630
557 if (maxreqs && !--maxreqs) 631 if (ecb_expect_false (maxreqs && !--maxreqs))
558 break; 632 break;
559 633
560 if (etp.max_poll_time) 634 if (maxtime)
561 { 635 {
562 gettimeofday (&tv_now, 0); 636 gettimeofday (&tv_now, 0);
563 637
564 if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time) 638 if (tvdiff (&tv_start, &tv_now) >= maxtime)
565 break; 639 break;
566 } 640 }
567 } 641 }
568 642
569 errno = EAGAIN; 643 errno = EAGAIN;
570 return -1; 644 return -1;
571} 645}
572 646
647static void
648etp_cancel (ETP_REQ *req)
649{
650 req->cancelled = 1;
651
652 eio_grp_cancel (req);
653}
654
655static void
656etp_submit (ETP_REQ *req)
657{
658 req->pri -= ETP_PRI_MIN;
659
660 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
661 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
662
663 if (ecb_expect_false (req->type == EIO_GROUP))
664 {
665 /* I hope this is worth it :/ */
666 X_LOCK (reqlock);
667 ++nreqs;
668 X_UNLOCK (reqlock);
669
670 X_LOCK (reslock);
671
672 ++npending;
673
674 if (!reqq_push (&res_queue, req) && want_poll_cb)
675 want_poll_cb ();
676
677 X_UNLOCK (reslock);
678 }
679 else
680 {
681 X_LOCK (reqlock);
682 ++nreqs;
683 ++nready;
684 reqq_push (&req_queue, req);
685 X_COND_SIGNAL (reqwait);
686 X_UNLOCK (reqlock);
687
688 etp_maybe_start_thread ();
689 }
690}
691
692static void ecb_cold
693etp_set_max_poll_time (double nseconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
696 max_poll_time = nseconds * EIO_TICKS;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
698}
699
700static void ecb_cold
701etp_set_max_poll_reqs (unsigned int maxreqs)
702{
703 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
704 max_poll_reqs = maxreqs;
705 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
706}
707
708static void ecb_cold
709etp_set_max_idle (unsigned int nthreads)
710{
711 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
712 max_idle = nthreads;
713 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
714}
715
716static void ecb_cold
717etp_set_idle_timeout (unsigned int seconds)
718{
719 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
720 idle_timeout = seconds;
721 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
722}
723
724static void ecb_cold
725etp_set_min_parallel (unsigned int nthreads)
726{
727 if (wanted < nthreads)
728 wanted = nthreads;
729}
730
731static void ecb_cold
732etp_set_max_parallel (unsigned int nthreads)
733{
734 if (wanted > nthreads)
735 wanted = nthreads;
736
737 while (started > wanted)
738 etp_end_thread ();
739}
740
741/*****************************************************************************/
742
743static void
744grp_try_feed (eio_req *grp)
745{
746 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
747 {
748 grp->flags &= ~EIO_FLAG_GROUPADD;
749
750 EIO_FEED (grp);
751
752 /* stop if no progress has been made */
753 if (!(grp->flags & EIO_FLAG_GROUPADD))
754 {
755 grp->feed = 0;
756 break;
757 }
758 }
759}
760
761static int
762grp_dec (eio_req *grp)
763{
764 --grp->size;
765
766 /* call feeder, if applicable */
767 grp_try_feed (grp);
768
769 /* finish, if done */
770 if (!grp->size && grp->int1)
771 return eio_finish (grp);
772 else
773 return 0;
774}
775
776static void
777eio_destroy (eio_req *req)
778{
779 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
780 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
781
782 EIO_DESTROY (req);
783}
784
785static int
786eio_finish (eio_req *req)
787{
788 int res = EIO_FINISH (req);
789
790 if (req->grp)
791 {
792 int res2;
793 eio_req *grp = req->grp;
794
795 /* unlink request */
796 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
797 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
798
799 if (grp->grp_first == req)
800 grp->grp_first = req->grp_next;
801
802 res2 = grp_dec (grp);
803
804 if (!res)
805 res = res2;
806 }
807
808 eio_destroy (req);
809
810 return res;
811}
812
813void
814eio_grp_cancel (eio_req *grp)
815{
816 for (grp = grp->grp_first; grp; grp = grp->grp_next)
817 eio_cancel (grp);
818}
819
820void
821eio_cancel (eio_req *req)
822{
823 etp_cancel (req);
824}
825
826void
827eio_submit (eio_req *req)
828{
829 etp_submit (req);
830}
831
832unsigned int
833eio_nreqs (void)
834{
835 return etp_nreqs ();
836}
837
838unsigned int
839eio_nready (void)
840{
841 return etp_nready ();
842}
843
844unsigned int
845eio_npending (void)
846{
847 return etp_npending ();
848}
849
850unsigned int ecb_cold
851eio_nthreads (void)
852{
853 return etp_nthreads ();
854}
855
856void ecb_cold
857eio_set_max_poll_time (double nseconds)
858{
859 etp_set_max_poll_time (nseconds);
860}
861
862void ecb_cold
863eio_set_max_poll_reqs (unsigned int maxreqs)
864{
865 etp_set_max_poll_reqs (maxreqs);
866}
867
868void ecb_cold
869eio_set_max_idle (unsigned int nthreads)
870{
871 etp_set_max_idle (nthreads);
872}
873
874void ecb_cold
875eio_set_idle_timeout (unsigned int seconds)
876{
877 etp_set_idle_timeout (seconds);
878}
879
880void ecb_cold
881eio_set_min_parallel (unsigned int nthreads)
882{
883 etp_set_min_parallel (nthreads);
884}
885
886void ecb_cold
887eio_set_max_parallel (unsigned int nthreads)
888{
889 etp_set_max_parallel (nthreads);
890}
891
892int eio_poll (void)
893{
894 return etp_poll ();
895}
896
573/*****************************************************************************/ 897/*****************************************************************************/
574/* work around various missing functions */ 898/* work around various missing functions */
575 899
576#if !HAVE_PREADWRITE 900#if !HAVE_PREADWRITE
901# undef pread
902# undef pwrite
577# define pread eio__pread 903# define pread eio__pread
578# define pwrite eio__pwrite 904# define pwrite eio__pwrite
579 905
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 906static eio_ssize_t
588eio__pread (int fd, void *buf, size_t count, off_t offset) 907eio__pread (int fd, void *buf, size_t count, off_t offset)
589{ 908{
590 ssize_t res; 909 eio_ssize_t res;
591 off_t ooffset; 910 off_t ooffset;
592 911
593 X_LOCK (preadwritelock); 912 X_LOCK (preadwritelock);
594 ooffset = lseek (fd, 0, SEEK_CUR); 913 ooffset = lseek (fd, 0, SEEK_CUR);
595 lseek (fd, offset, SEEK_SET); 914 lseek (fd, offset, SEEK_SET);
… …
598 X_UNLOCK (preadwritelock); 917 X_UNLOCK (preadwritelock);
599 918
600 return res; 919 return res;
601} 920}
602 921
603static ssize_t 922static eio_ssize_t
604eio__pwrite (int fd, void *buf, size_t count, off_t offset) 923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
605{ 924{
606 ssize_t res; 925 eio_ssize_t res;
607 off_t ooffset; 926 off_t ooffset;
608 927
609 X_LOCK (preadwritelock); 928 X_LOCK (preadwritelock);
610 ooffset = lseek (fd, 0, SEEK_CUR); 929 ooffset = lseek (fd, 0, SEEK_CUR);
611 lseek (fd, offset, SEEK_SET); 930 lseek (fd, offset, SEEK_SET);
612 res = write (fd, buf, count); 931 res = write (fd, buf, count);
613 lseek (fd, offset, SEEK_SET); 932 lseek (fd, ooffset, SEEK_SET);
614 X_UNLOCK (preadwritelock); 933 X_UNLOCK (preadwritelock);
615 934
616 return res; 935 return res;
617} 936}
618#endif 937#endif
619 938
620#ifndef HAVE_FUTIMES 939#ifndef HAVE_UTIMES
621 940
941# undef utimes
622# define utimes(path,times) eio__utimes (path, times) 942# define utimes(path,times) eio__utimes (path, times)
623# define futimes(fd,times) eio__futimes (fd, times)
624 943
625static int 944static int
626eio__utimes (const char *filename, const struct timeval times[2]) 945eio__utimes (const char *filename, const struct timeval times[2])
627{ 946{
628 if (times) 947 if (times)
… …
636 } 955 }
637 else 956 else
638 return utime (filename, 0); 957 return utime (filename, 0);
639} 958}
640 959
960#endif
961
962#ifndef HAVE_FUTIMES
963
964# undef futimes
965# define futimes(fd,times) eio__futimes (fd, times)
966
967static int
641static int eio__futimes (int fd, const struct timeval tv[2]) 968eio__futimes (int fd, const struct timeval tv[2])
642{ 969{
643 errno = ENOSYS; 970 errno = ENOSYS;
644 return -1; 971 return -1;
645} 972}
646 973
647#endif 974#endif
648 975
649#if !HAVE_FDATASYNC 976#if !HAVE_FDATASYNC
977# undef fdatasync
650# define fdatasync fsync 978# define fdatasync(fd) fsync (fd)
651#endif 979#endif
980
981static int
982eio__syncfs (int fd)
983{
984 int res;
985
986#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else
989 res = -1;
990 errno = ENOSYS;
991#endif
992
993 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync ();
995
996 return res;
997}
998
999/* sync_file_range always needs emulation */
1000static int
1001eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1002{
1003#if HAVE_SYNC_FILE_RANGE
1004 int res;
1005
1006 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1007 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1008 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1009 {
1010 flags = 0
1011 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1012 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1013 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1014 }
1015
1016 res = sync_file_range (fd, offset, nbytes, flags);
1017
1018 if (!res || errno != ENOSYS)
1019 return res;
1020#endif
1021
1022 /* even though we could play tricks with the flags, it's better to always
1023 * call fdatasync, as that matches the expectation of its users best */
1024 return fdatasync (fd);
1025}
1026
1027static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{
1030#if HAVE_FALLOCATE
1031 return fallocate (fd, mode, offset, len);
1032#else
1033 errno = ENOSYS;
1034 return -1;
1035#endif
1036}
652 1037
653#if !HAVE_READAHEAD 1038#if !HAVE_READAHEAD
1039# undef readahead
654# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1040# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655 1041
656static ssize_t 1042static eio_ssize_t
657eio__readahead (int fd, off_t offset, size_t count, worker *self) 1043eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658{ 1044{
659 size_t todo = count; 1045 size_t todo = count;
660 dBUF; 1046 dBUF;
661 1047
662 while (todo > 0) 1048 while (todo > 0)
… …
666 pread (fd, eio_buf, len, offset); 1052 pread (fd, eio_buf, len, offset);
667 offset += len; 1053 offset += len;
668 todo -= len; 1054 todo -= len;
669 } 1055 }
670 1056
1057 FUBd;
1058
671 errno = 0; 1059 errno = 0;
672 return count; 1060 return count;
673} 1061}
674 1062
675#endif 1063#endif
676 1064
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 */ 1065/* sendfile always needs emulation */
709static ssize_t 1066static eio_ssize_t
710eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 1067eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
711{ 1068{
1069 eio_ssize_t written = 0;
712 ssize_t res; 1070 eio_ssize_t res;
713 1071
714 if (!count) 1072 if (!count)
715 return 0; 1073 return 0;
716 1074
1075 for (;;)
1076 {
1077#ifdef __APPLE__
1078# undef HAVE_SENDFILE /* broken, as everything on os x */
1079#endif
717#if HAVE_SENDFILE 1080#if HAVE_SENDFILE
718# if __linux 1081# if __linux
1082 off_t soffset = offset;
719 res = sendfile (ofd, ifd, &offset, count); 1083 res = sendfile (ofd, ifd, &soffset, count);
720 1084
721# elif __freebsd 1085# elif __FreeBSD__
722 /* 1086 /*
723 * Of course, the freebsd sendfile is a dire hack with no thoughts 1087 * Of course, the freebsd sendfile is a dire hack with no thoughts
724 * wasted on making it similar to other I/O functions. 1088 * wasted on making it similar to other I/O functions.
725 */ 1089 */
726 {
727 off_t sbytes; 1090 off_t sbytes;
728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1091 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729 1092
730 if (res < 0 && sbytes) 1093 #if 0 /* according to the manpage, this is correct, but broken behaviour */
731 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1094 /* freebsd' sendfile will return 0 on success */
1095 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1096 /* not on e.g. EIO or EPIPE - sounds broken */
1097 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
732 res = sbytes; 1098 res = sbytes;
733 } 1099 #endif
1100
1101 /* according to source inspection, this is correct, and useful behaviour */
1102 if (sbytes)
1103 res = sbytes;
1104
1105# elif defined (__APPLE__)
1106 off_t sbytes = count;
1107 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1108
1109 /* according to the manpage, sbytes is always valid */
1110 if (sbytes)
1111 res = sbytes;
734 1112
735# elif __hpux 1113# elif __hpux
736 res = sendfile (ofd, ifd, offset, count, 0, 0); 1114 res = sendfile (ofd, ifd, offset, count, 0, 0);
737 1115
738# elif __solaris 1116# elif __solaris
739 {
740 struct sendfilevec vec; 1117 struct sendfilevec vec;
741 size_t sbytes; 1118 size_t sbytes;
742 1119
743 vec.sfv_fd = ifd; 1120 vec.sfv_fd = ifd;
744 vec.sfv_flag = 0; 1121 vec.sfv_flag = 0;
745 vec.sfv_off = offset; 1122 vec.sfv_off = offset;
746 vec.sfv_len = count; 1123 vec.sfv_len = count;
747 1124
748 res = sendfilev (ofd, &vec, 1, &sbytes); 1125 res = sendfilev (ofd, &vec, 1, &sbytes);
749 1126
750 if (res < 0 && sbytes) 1127 if (res < 0 && sbytes)
751 res = sbytes; 1128 res = sbytes;
752 }
753 1129
754# endif 1130# endif
1131
1132#elif defined (_WIN32) && 0
1133 /* does not work, just for documentation of what would need to be done */
1134 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1135 /* libeio guarantees that the file offset does not change, and windows */
1136 /* has no way to get an independent handle to the same file description */
1137 HANDLE h = TO_SOCKET (ifd);
1138 SetFilePointer (h, offset, 0, FILE_BEGIN);
1139 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1140
755#else 1141#else
756 res = -1; 1142 res = -1;
757 errno = ENOSYS; 1143 errno = ENOSYS;
758#endif 1144#endif
759 1145
1146 /* we assume sendfile can copy at least 128mb in one go */
1147 if (res <= 128 * 1024 * 1024)
1148 {
1149 if (res > 0)
1150 written += res;
1151
1152 if (written)
1153 return written;
1154
1155 break;
1156 }
1157 else
1158 {
1159 /* if we requested more, then probably the kernel was lazy */
1160 written += res;
1161 offset += res;
1162 count -= res;
1163
1164 if (!count)
1165 return written;
1166 }
1167 }
1168
760 if (res < 0 1169 if (res < 0
761 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1170 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1171 /* BSDs */
1172#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1173 || errno == ENOTSUP
1174#endif
1175#ifdef EOPNOTSUPP /* windows */
1176 || errno == EOPNOTSUPP /* BSDs */
1177#endif
762#if __solaris 1178#if __solaris
763 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1179 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
764#endif 1180#endif
765 ) 1181 )
766 ) 1182 )
… …
770 1186
771 res = 0; 1187 res = 0;
772 1188
773 while (count) 1189 while (count)
774 { 1190 {
775 ssize_t cnt; 1191 eio_ssize_t cnt;
776 1192
777 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1193 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
778 1194
779 if (cnt <= 0) 1195 if (cnt <= 0)
780 { 1196 {
… …
792 1208
793 offset += cnt; 1209 offset += cnt;
794 res += cnt; 1210 res += cnt;
795 count -= cnt; 1211 count -= cnt;
796 } 1212 }
1213
1214 FUBd;
797 } 1215 }
798 1216
799 return res; 1217 return res;
1218}
1219
1220#ifdef PAGESIZE
1221# define eio_pagesize() PAGESIZE
1222#else
1223static intptr_t
1224eio_pagesize (void)
1225{
1226 static intptr_t page;
1227
1228 if (!page)
1229 page = sysconf (_SC_PAGESIZE);
1230
1231 return page;
1232}
1233#endif
1234
1235static void
1236eio_page_align (void **addr, size_t *length)
1237{
1238 intptr_t mask = eio_pagesize () - 1;
1239
1240 /* round down addr */
1241 intptr_t adj = mask & (intptr_t)*addr;
1242
1243 *addr = (void *)((intptr_t)*addr - adj);
1244 *length += adj;
1245
1246 /* round up length */
1247 *length = (*length + mask) & ~mask;
1248}
1249
1250#if !_POSIX_MEMLOCK
1251# define eio__mlockall(a) EIO_ENOSYS ()
1252#else
1253
1254static int
1255eio__mlockall (int flags)
1256{
1257 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1258 extern int mallopt (int, int);
1259 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1260 #endif
1261
1262 if (EIO_MCL_CURRENT != MCL_CURRENT
1263 || EIO_MCL_FUTURE != MCL_FUTURE)
1264 {
1265 flags = 0
1266 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1267 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1268 }
1269
1270 return mlockall (flags);
1271}
1272#endif
1273
1274#if !_POSIX_MEMLOCK_RANGE
1275# define eio__mlock(a,b) EIO_ENOSYS ()
1276#else
1277
1278static int
1279eio__mlock (void *addr, size_t length)
1280{
1281 eio_page_align (&addr, &length);
1282
1283 return mlock (addr, length);
1284}
1285
1286#endif
1287
1288#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1289# define eio__msync(a,b,c) EIO_ENOSYS ()
1290#else
1291
1292static int
1293eio__msync (void *mem, size_t len, int flags)
1294{
1295 eio_page_align (&mem, &len);
1296
1297 if (EIO_MS_ASYNC != MS_SYNC
1298 || EIO_MS_INVALIDATE != MS_INVALIDATE
1299 || EIO_MS_SYNC != MS_SYNC)
1300 {
1301 flags = 0
1302 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1303 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1304 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1305 }
1306
1307 return msync (mem, len, flags);
1308}
1309
1310#endif
1311
1312static int
1313eio__mtouch (eio_req *req)
1314{
1315 void *mem = req->ptr2;
1316 size_t len = req->size;
1317 int flags = req->int1;
1318
1319 eio_page_align (&mem, &len);
1320
1321 {
1322 intptr_t addr = (intptr_t)mem;
1323 intptr_t end = addr + len;
1324 intptr_t page = eio_pagesize ();
1325
1326 if (addr < end)
1327 if (flags & EIO_MT_MODIFY) /* modify */
1328 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1329 else
1330 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1331 }
1332
1333 return 0;
1334}
1335
1336/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */
1338
1339static void
1340eio__realpath (eio_req *req, etp_worker *self)
1341{
1342 char *rel = req->ptr1;
1343 char *res;
1344 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX;
1347#else
1348 int symlinks = 32;
1349#endif
1350
1351 req->result = -1;
1352
1353 errno = EINVAL;
1354 if (!rel)
1355 return;
1356
1357 errno = ENOENT;
1358 if (!*rel)
1359 return;
1360
1361 if (!req->ptr2)
1362 {
1363 X_LOCK (wrklock);
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return;
1371 }
1372
1373 res = req->ptr2;
1374 tmp1 = res + PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX;
1376
1377#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1379 /* on linux we may be able to ask the kernel */
1380 {
1381 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1382
1383 if (fd >= 0)
1384 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX);
1387 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */
1390
1391 if (req->result > 0)
1392 goto done;
1393 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return;
1396 }
1397#endif
1398#endif
1399
1400 if (*rel != '/')
1401 {
1402 if (!getcwd (res, PATH_MAX))
1403 return;
1404
1405 if (res [1]) /* only use if not / */
1406 res += strlen (res);
1407 }
1408
1409 while (*rel)
1410 {
1411 eio_ssize_t len, linklen;
1412 char *beg = rel;
1413
1414 while (*rel && *rel != '/')
1415 ++rel;
1416
1417 len = rel - beg;
1418
1419 if (!len) /* skip slashes */
1420 {
1421 ++rel;
1422 continue;
1423 }
1424
1425 if (beg [0] == '.')
1426 {
1427 if (len == 1)
1428 continue; /* . - nop */
1429
1430 if (beg [1] == '.' && len == 2)
1431 {
1432 /* .. - back up one component, if possible */
1433
1434 while (res != req->ptr2)
1435 if (*--res == '/')
1436 break;
1437
1438 continue;
1439 }
1440 }
1441
1442 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1)
1444 return;
1445
1446 /* copy one component */
1447 *res = '/';
1448 memcpy (res + 1, beg, len);
1449
1450 /* zero-terminate, for readlink */
1451 res [len + 1] = 0;
1452
1453 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1455
1456 if (linklen < 0)
1457 {
1458 if (errno != EINVAL)
1459 return;
1460
1461 /* it's a normal directory. hopefully */
1462 res += len + 1;
1463 }
1464 else
1465 {
1466 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel);
1468
1469 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX)
1471 return;
1472
1473 errno = ELOOP;
1474 if (!--symlinks)
1475 return;
1476
1477 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */
1479
1480 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen);
1484
1485 rel = tmp2;
1486 }
1487 }
1488
1489 /* special case for the lone root path */
1490 if (res == req->ptr2)
1491 *res++ = '/';
1492
1493 req->result = res - (char *)req->ptr2;
1494
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1497}
1498
1499static signed char
1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1501{
1502 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1503 : a->inode < b->inode ? -1
1504 : a->inode > b->inode ? 1
1505 : 0;
1506}
1507
1508#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1509
1510#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1511#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1512
1513static void
1514eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1515{
1516 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1517 unsigned char *bit = bits;
1518
1519 assert (CHAR_BIT == 8);
1520 assert (sizeof (eio_dirent) * 8 < 256);
1521 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1522 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1523
1524 if (size <= EIO_SORT_FAST)
1525 return;
1526
1527 /* first prepare an array of bits to test in our radix sort */
1528 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1529 /* inode_bits must contain all inodes ORed together */
1530 /* which is used to skip bits that are 0 everywhere, which is very common */
1531 {
1532 eio_ino_t endianness;
1533 int i, j;
1534
1535 /* we store the byte offset of byte n into byte n of "endianness" */
1536 for (i = 0; i < sizeof (eio_ino_t); ++i)
1537 ((unsigned char *)&endianness)[i] = i;
1538
1539 *bit++ = 0;
1540
1541 for (i = 0; i < sizeof (eio_ino_t); ++i)
1542 {
1543 /* shifting off the byte offsets out of "endianness" */
1544 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1545 endianness >>= 8;
1546
1547 for (j = 0; j < 8; ++j)
1548 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1549 *bit++ = offs + j;
1550 }
1551
1552 for (j = 0; j < 8; ++j)
1553 if (score_bits & (1 << j))
1554 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1555 }
1556
1557 /* now actually do the sorting (a variant of MSD radix sort) */
1558 {
1559 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1560 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1561 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1562 int stk_idx = 0;
1563
1564 base_stk [stk_idx] = dents;
1565 end_stk [stk_idx] = dents + size;
1566 bit_stk [stk_idx] = bit - 1;
1567
1568 do
1569 {
1570 base = base_stk [stk_idx];
1571 end = end_stk [stk_idx];
1572 bit = bit_stk [stk_idx];
1573
1574 for (;;)
1575 {
1576 unsigned char O = *bit >> 3;
1577 unsigned char M = 1 << (*bit & 7);
1578
1579 eio_dirent *a = base;
1580 eio_dirent *b = end;
1581
1582 if (b - a < EIO_SORT_CUTOFF)
1583 break;
1584
1585 /* now bit-partition the array on the bit */
1586 /* this ugly asymmetric loop seems to perform much better than typical */
1587 /* partition algos found in the literature */
1588 do
1589 if (!(((unsigned char *)a)[O] & M))
1590 ++a;
1591 else if (!(((unsigned char *)--b)[O] & M))
1592 {
1593 eio_dirent tmp = *a; *a = *b; *b = tmp;
1594 ++a;
1595 }
1596 while (b > a);
1597
1598 /* next bit, or stop, if no bits left in this path */
1599 if (!*--bit)
1600 break;
1601
1602 base_stk [stk_idx] = a;
1603 end_stk [stk_idx] = end;
1604 bit_stk [stk_idx] = bit;
1605 ++stk_idx;
1606
1607 end = a;
1608 }
1609 }
1610 while (stk_idx--);
1611 }
1612}
1613
1614static void
1615eio_dent_insertion_sort (eio_dirent *dents, int size)
1616{
1617 /* first move the smallest element to the front, to act as a sentinel */
1618 {
1619 int i;
1620 eio_dirent *min = dents;
1621
1622 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1623 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1624 if (EIO_DENT_CMP (dents [i], <, *min))
1625 min = &dents [i];
1626
1627 /* swap elements 0 and j (minimum) */
1628 {
1629 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1630 }
1631 }
1632
1633 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1634 {
1635 eio_dirent *i, *j;
1636
1637 for (i = dents + 1; i < dents + size; ++i)
1638 {
1639 eio_dirent value = *i;
1640
1641 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1642 j [1] = j [0];
1643
1644 j [1] = value;
1645 }
1646 }
1647}
1648
1649static void
1650eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1651{
1652 if (size <= 1)
1653 return; /* our insertion sort relies on size > 0 */
1654
1655 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1656 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1657 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1658
1659 /* use an insertion sort at the end, or for small arrays, */
1660 /* as insertion sort is more efficient for small partitions */
1661 eio_dent_insertion_sort (dents, size);
800} 1662}
801 1663
802/* read a full directory */ 1664/* read a full directory */
803static void 1665static void
804eio__scandir (eio_req *req, worker *self) 1666eio__scandir (eio_req *req, etp_worker *self)
805{ 1667{
1668 char *name, *names;
1669 int namesalloc = 4096 - sizeof (void *) * 4;
1670 int namesoffs = 0;
1671 int flags = req->int1;
1672 eio_dirent *dents = 0;
1673 int dentalloc = 128;
1674 int dentoffs = 0;
1675 eio_ino_t inode_bits = 0;
1676#ifdef _WIN32
1677 HANDLE dirp;
1678 WIN32_FIND_DATA entp;
1679#else
806 DIR *dirp; 1680 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; 1681 EIO_STRUCT_DIRENT *entp;
813 char *name, *names; 1682#endif
814 int memlen = 4096;
815 int memofs = 0;
816 int res = 0;
817 1683
818 X_LOCK (etplock); 1684 req->result = -1;
1685
1686 if (!(flags & EIO_READDIR_DENTS))
1687 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1688
1689#ifdef _WIN32
1690 {
1691 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH);
1693 const char *fmt;
1694
1695 if (!len)
1696 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1698 fmt = "%s*";
1699 else
1700 fmt = "%s/*";
1701
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1703 dirp = FindFirstFile (path, &entp);
1704 free (path);
1705
1706 if (dirp == INVALID_HANDLE_VALUE)
1707 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */
1711 switch (GetLastError ())
1712 {
1713 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0;
1715 break;
1716
1717 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT;
1721 break;
1722
1723 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM;
1725 break;
1726
1727 default:
1728 errno = EINVAL;
1729 break;
1730 }
1731 }
1732 }
1733#else
819 self->dirp = dirp = opendir (req->ptr1); 1734 dirp = opendir (req->ptr1);
820 self->dbuf = u = malloc (sizeof (*u)); 1735#endif
1736
821 req->flags |= EIO_FLAG_PTR2_FREE; 1737 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1);
1739
1740 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822 req->ptr2 = names = malloc (memlen); 1742 req->ptr2 = names = malloc (namesalloc);
823 X_UNLOCK (etplock);
824 1743
825 if (dirp && u && names) 1744 if (dirp && names && (!flags || dents))
826 for (;;) 1745 for (;;)
827 { 1746 {
1747 int done;
1748
1749#ifdef _WIN32
1750 done = !dirp;
1751#else
828 errno = 0; 1752 errno = 0;
829 readdir_r (dirp, &u->d, &entp); 1753 entp = readdir (dirp);
1754 done = !entp;
1755#endif
830 1756
831 if (!entp) 1757 if (done)
1758 {
1759#ifndef _WIN32
1760 int old_errno = errno;
1761 closedir (dirp);
1762 errno = old_errno;
1763
1764 if (errno)
1765 break;
1766#endif
1767
1768 /* sort etc. */
1769 req->int1 = flags;
1770 req->result = dentoffs;
1771
1772 if (flags & EIO_READDIR_STAT_ORDER)
1773 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1774 else if (flags & EIO_READDIR_DIRS_FIRST)
1775 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1776 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1777 else
1778 {
1779 /* in this case, all is known, and we just put dirs first and sort them */
1780 eio_dirent *oth = dents + dentoffs;
1781 eio_dirent *dir = dents;
1782
1783 /* now partition dirs to the front, and non-dirs to the back */
1784 /* by walking from both sides and swapping if necessary */
1785 while (oth > dir)
1786 {
1787 if (dir->type == EIO_DT_DIR)
1788 ++dir;
1789 else if ((--oth)->type == EIO_DT_DIR)
1790 {
1791 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1792
1793 ++dir;
1794 }
1795 }
1796
1797 /* now sort the dirs only (dirs all have the same score) */
1798 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1799 }
1800
832 break; 1801 break;
1802 }
833 1803
834 name = entp->d_name; 1804 /* now add the entry to our list(s) */
1805 name = D_NAME (entp);
835 1806
1807 /* skip . and .. entries */
836 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837 { 1809 {
838 int len = strlen (name) + 1; 1810 int len = D_NAMLEN (entp) + 1;
839 1811
840 res++; 1812 while (ecb_expect_false (namesoffs + len > namesalloc))
841
842 while (memofs + len > memlen)
843 { 1813 {
844 memlen *= 2; 1814 namesalloc *= 2;
845 X_LOCK (etplock);
846 req->ptr2 = names = realloc (names, memlen); 1815 req->ptr2 = names = realloc (names, namesalloc);
847 X_UNLOCK (etplock);
848 1816
849 if (!names) 1817 if (!names)
850 break; 1818 break;
851 } 1819 }
852 1820
853 memcpy (names + memofs, name, len); 1821 memcpy (names + namesoffs, name, len);
1822
1823 if (dents)
1824 {
1825 struct eio_dirent *ent;
1826
1827 if (ecb_expect_false (dentoffs == dentalloc))
1828 {
1829 dentalloc *= 2;
1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1831
1832 if (!dents)
1833 break;
1834 }
1835
1836 ent = dents + dentoffs;
1837
1838 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1839 ent->namelen = len - 1;
1840 ent->inode = D_INO (entp);
1841
1842 inode_bits |= ent->inode;
1843
1844 switch (D_TYPE (entp))
1845 {
1846 default:
1847 ent->type = EIO_DT_UNKNOWN;
1848 flags |= EIO_READDIR_FOUND_UNKNOWN;
1849 break;
1850
1851 #ifdef DT_FIFO
1852 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1853 #endif
1854 #ifdef DT_CHR
1855 case DT_CHR: ent->type = EIO_DT_CHR; break;
1856 #endif
1857 #ifdef DT_MPC
1858 case DT_MPC: ent->type = EIO_DT_MPC; break;
1859 #endif
1860 #ifdef DT_DIR
1861 case DT_DIR: ent->type = EIO_DT_DIR; break;
1862 #endif
1863 #ifdef DT_NAM
1864 case DT_NAM: ent->type = EIO_DT_NAM; break;
1865 #endif
1866 #ifdef DT_BLK
1867 case DT_BLK: ent->type = EIO_DT_BLK; break;
1868 #endif
1869 #ifdef DT_MPB
1870 case DT_MPB: ent->type = EIO_DT_MPB; break;
1871 #endif
1872 #ifdef DT_REG
1873 case DT_REG: ent->type = EIO_DT_REG; break;
1874 #endif
1875 #ifdef DT_NWK
1876 case DT_NWK: ent->type = EIO_DT_NWK; break;
1877 #endif
1878 #ifdef DT_CMP
1879 case DT_CMP: ent->type = EIO_DT_CMP; break;
1880 #endif
1881 #ifdef DT_LNK
1882 case DT_LNK: ent->type = EIO_DT_LNK; break;
1883 #endif
1884 #ifdef DT_SOCK
1885 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1886 #endif
1887 #ifdef DT_DOOR
1888 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1889 #endif
1890 #ifdef DT_WHT
1891 case DT_WHT: ent->type = EIO_DT_WHT; break;
1892 #endif
1893 }
1894
1895 ent->score = 7;
1896
1897 if (flags & EIO_READDIR_DIRS_FIRST)
1898 {
1899 if (ent->type == EIO_DT_UNKNOWN)
1900 {
1901 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1902 ent->score = 1;
1903 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1904 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1905 }
1906 else if (ent->type == EIO_DT_DIR)
1907 ent->score = 0;
1908 }
1909 }
1910
854 memofs += len; 1911 namesoffs += len;
1912 ++dentoffs;
855 } 1913 }
1914
1915 if (EIO_CANCELLED (req))
1916 {
1917 errno = ECANCELED;
1918 break;
1919 }
1920
1921#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp))
1923 {
1924 FindClose (dirp);
1925 dirp = 0;
1926 }
1927#endif
856 } 1928 }
857
858 if (errno)
859 res = -1;
860
861 req->result = res;
862} 1929}
863 1930
864/*****************************************************************************/ 1931/*****************************************************************************/
865 1932
866#define ALLOC(len) \ 1933#define ALLOC(len) \
867 if (!req->ptr2) \ 1934 if (!req->ptr2) \
868 { \ 1935 { \
869 X_LOCK (etplock); \ 1936 X_LOCK (wrklock); \
870 req->flags |= EIO_FLAG_PTR2_FREE; \ 1937 req->flags |= EIO_FLAG_PTR2_FREE; \
871 X_UNLOCK (etplock); \ 1938 X_UNLOCK (wrklock); \
872 req->ptr2 = malloc (len); \ 1939 req->ptr2 = malloc (len); \
873 if (!req->ptr2) \ 1940 if (!req->ptr2) \
874 { \ 1941 { \
875 errno = ENOMEM; \ 1942 errno = ENOMEM; \
876 req->result = -1; \ 1943 req->result = -1; \
877 break; \ 1944 break; \
878 } \ 1945 } \
879 } 1946 }
880 1947
881X_THREAD_PROC (eio_proc) 1948X_THREAD_PROC (etp_proc)
882{ 1949{
883 eio_req *req; 1950 ETP_REQ *req;
884 struct timespec ts; 1951 struct timespec ts;
885 worker *self = (worker *)thr_arg; 1952 etp_worker *self = (etp_worker *)thr_arg;
886 1953
1954#if HAVE_PRCTL_SET_NAME
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1956#endif
1957
887 /* try to distribute timeouts somewhat randomly */ 1958 /* try to distribute timeouts somewhat evenly */
888 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889 1960
890 for (;;) 1961 for (;;)
891 { 1962 {
892 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1963 ts.tv_sec = 0;
893 1964
894 X_LOCK (reqlock); 1965 X_LOCK (reqlock);
895 1966
896 for (;;) 1967 for (;;)
897 { 1968 {
898 self->req = req = reqq_shift (&req_queue); 1969 self->req = req = reqq_shift (&req_queue);
899 1970
900 if (req) 1971 if (req)
901 break; 1972 break;
902 1973
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1975 {
1976 X_UNLOCK (reqlock);
1977 X_LOCK (wrklock);
1978 --started;
1979 X_UNLOCK (wrklock);
1980 goto quit;
1981 }
1982
903 ++idle; 1983 ++idle;
904 1984
905 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1985 if (idle <= max_idle)
1986 /* we are allowed to idle, so do so without any timeout */
1987 X_COND_WAIT (reqwait, reqlock);
1988 else
906 { 1989 {
907 if (idle > max_idle) 1990 /* initialise timeout once */
908 { 1991 if (!ts.tv_sec)
909 --idle; 1992 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 1993
917 /* we are allowed to idle, so do so without any timeout */
918 X_COND_WAIT (reqwait, reqlock); 1994 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
919 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1995 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
920 } 1996 }
921 1997
922 --idle; 1998 --idle;
923 } 1999 }
924 2000
925 --nready; 2001 --nready;
926 2002
927 X_UNLOCK (reqlock); 2003 X_UNLOCK (reqlock);
928 2004
929 errno = 0; /* strictly unnecessary */
930
931 if (!EIO_CANCELLED (req))
932 switch (req->type) 2005 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; 2006 goto quit;
1025 2007
1026 default: 2008 ETP_EXECUTE (self, req);
1027 req->result = -1;
1028 break;
1029 }
1030
1031 req->errorno = errno;
1032 2009
1033 X_LOCK (reslock); 2010 X_LOCK (reslock);
1034 2011
1035 ++npending; 2012 ++npending;
1036 2013
1037 if (!reqq_push (&res_queue, req) && etp.want_poll_cb) 2014 if (!reqq_push (&res_queue, req) && want_poll_cb)
1038 etp.want_poll_cb (); 2015 want_poll_cb ();
1039 2016
1040 self->req = 0; 2017 self->req = 0;
1041 worker_clear (self); 2018 etp_worker_clear (self);
1042 2019
1043 X_UNLOCK (reslock); 2020 X_UNLOCK (reslock);
1044 } 2021 }
1045 2022
1046quit: 2023quit:
1047 X_LOCK (etplock); 2024 X_LOCK (wrklock);
1048 worker_free (self); 2025 etp_worker_free (self);
1049 X_UNLOCK (etplock); 2026 X_UNLOCK (wrklock);
1050 2027
1051 return 0; 2028 return 0;
1052} 2029}
1053 2030
1054/*****************************************************************************/ 2031/*****************************************************************************/
1055 2032
2033int ecb_cold
1056int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2034eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057{ 2035{
2036#if !HAVE_PREADWRITE
2037 X_MUTEX_CREATE (preadwritelock);
2038#endif
2039
1058 etp_init (&etp, want_poll, done_poll); 2040 return etp_init (want_poll, done_poll);
1059} 2041}
1060 2042
2043ecb_inline void
1061static void eio_api_destroy (eio_req *req) 2044eio_api_destroy (eio_req *req)
1062{ 2045{
1063 free (req); 2046 free (req);
1064} 2047}
1065 2048
1066#define REQ(rtype) \ 2049#define REQ(rtype) \
… …
1085 { \ 2068 { \
1086 eio_api_destroy (req); \ 2069 eio_api_destroy (req); \
1087 return 0; \ 2070 return 0; \
1088 } 2071 }
1089 2072
2073static void
2074eio_execute (etp_worker *self, eio_req *req)
2075{
2076 if (ecb_expect_false (EIO_CANCELLED (req)))
2077 {
2078 req->result = -1;
2079 req->errorno = ECANCELED;
2080 return;
2081 }
2082
2083 switch (req->type)
2084 {
2085 case EIO_READ: ALLOC (req->size);
2086 req->result = req->offs >= 0
2087 ? pread (req->int1, req->ptr2, req->size, req->offs)
2088 : read (req->int1, req->ptr2, req->size); break;
2089 case EIO_WRITE: req->result = req->offs >= 0
2090 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2091 : write (req->int1, req->ptr2, req->size); break;
2092
2093 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2095
2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2102
2103 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2104 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2105 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2106 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2107
2108 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2109 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2110 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2111 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2112 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2113 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2114
2115 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2116 case EIO_CLOSE: req->result = close (req->int1); break;
2117 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2118 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2119 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2120 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
2128 case EIO_READLINK: ALLOC (PATH_MAX);
2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2130
2131 case EIO_SYNC: req->result = 0; sync (); break;
2132 case EIO_FSYNC: req->result = fsync (req->int1); break;
2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2134 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2136 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2137 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2138 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2139 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2140 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2141
2142 case EIO_READDIR: eio__scandir (req, self); break;
2143
2144 case EIO_BUSY:
2145#ifdef _WIN32
2146 Sleep (req->nv1 * 1e3);
2147#else
2148 {
2149 struct timeval tv;
2150
2151 tv.tv_sec = req->nv1;
2152 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2153
2154 req->result = select (0, 0, 0, 0, &tv);
2155 }
2156#endif
2157 break;
2158
2159 case EIO_UTIME:
2160 case EIO_FUTIME:
2161 {
2162 struct timeval tv[2];
2163 struct timeval *times;
2164
2165 if (req->nv1 != -1. || req->nv2 != -1.)
2166 {
2167 tv[0].tv_sec = req->nv1;
2168 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2169 tv[1].tv_sec = req->nv2;
2170 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2171
2172 times = tv;
2173 }
2174 else
2175 times = 0;
2176
2177 req->result = req->type == EIO_FUTIME
2178 ? futimes (req->int1, times)
2179 : utimes (req->ptr1, times);
2180 }
2181 break;
2182
2183 case EIO_GROUP:
2184 abort (); /* handled in eio_request */
2185
2186 case EIO_NOP:
2187 req->result = 0;
2188 break;
2189
2190 case EIO_CUSTOM:
2191 req->feed (req);
2192 break;
2193
2194 default:
2195 errno = ENOSYS;
2196 req->result = -1;
2197 break;
2198 }
2199
2200 req->errorno = errno;
2201}
2202
1090#ifndef EIO_NO_WRAPPERS 2203#ifndef EIO_NO_WRAPPERS
1091 2204
1092eio_req *eio_nop (int pri, eio_cb cb, void *data) 2205eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093{ 2206{
1094 REQ (EIO_NOP); SEND; 2207 REQ (EIO_NOP); SEND;
… …
1107eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2220eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108{ 2221{
1109 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2222 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110} 2223}
1111 2224
2225eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2226{
2227 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2228}
2229
1112eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2230eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113{ 2231{
1114 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2232 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115} 2233}
1116 2234
2235eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2236{
2237 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2238}
2239
2240eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2243}
2244
2245eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2246{
2247 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2248}
2249
2250eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2251{
2252 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2253}
2254
2255eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2256{
2257 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2258}
2259
2260eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2261{
2262 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2263}
2264
1117eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2265eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1118{ 2266{
1119 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2267 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1120} 2268}
1121 2269
… …
1137eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2285eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138{ 2286{
1139 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2287 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1140} 2288}
1141 2289
2290eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2291{
2292 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2293}
2294
1142eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2295eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143{ 2296{
1144 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2297 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1145} 2298}
1146 2299
… …
1152eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2305eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2306{
1154 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2307 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1155} 2308}
1156 2309
1157eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2310eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1158{ 2311{
1159 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2312 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1160} 2313}
1161 2314
1162eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2315eio_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) 2335eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1183{ 2336{
1184 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2337 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1185} 2338}
1186 2339
1187eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2340eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1188{ 2341{
1189 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2342 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1190} 2343}
1191 2344
1192eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2345eio_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) 2361eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1209{ 2362{
1210 return eio__1path (EIO_READLINK, path, pri, cb, data); 2363 return eio__1path (EIO_READLINK, path, pri, cb, data);
1211} 2364}
1212 2365
2366eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2367{
2368 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2369}
2370
1213eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2371eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1214{ 2372{
1215 return eio__1path (EIO_STAT, path, pri, cb, data); 2373 return eio__1path (EIO_STAT, path, pri, cb, data);
1216} 2374}
1217 2375
1218eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2376eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219{ 2377{
1220 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2378 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221} 2379}
1222 2380
2381eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2382{
2383 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2384}
2385
1223eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2386eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224{ 2387{
1225 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2388 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226} 2389}
1227 2390
1228eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2391eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229{ 2392{
1230 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2393 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231} 2394}
1232 2395
1233eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2396eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234{ 2397{
1235 return eio__1path (EIO_READDIR, path, pri, cb, data); 2398 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236} 2399}
1237 2400
1238eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2401eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239{ 2402{
1240 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2403 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1241} 2404}
1242 2405
1243static eio_req * 2406static eio_req *
1244eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2407eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245{ 2408{
… …
1269eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2432eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270{ 2433{
1271 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2434 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272} 2435}
1273 2436
2437eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2438{
2439 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2440}
2441
1274#endif 2442#endif
1275 2443
1276eio_req *eio_grp (eio_cb cb, void *data) 2444eio_req *eio_grp (eio_cb cb, void *data)
1277{ 2445{
1278 const int pri = EIO_PRI_MAX; 2446 const int pri = EIO_PRI_MAX;
… …
1285#undef SEND 2453#undef SEND
1286 2454
1287/*****************************************************************************/ 2455/*****************************************************************************/
1288/* grp functions */ 2456/* grp functions */
1289 2457
2458void
1290void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2459eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1291{ 2460{
1292 grp->int2 = limit; 2461 grp->int2 = limit;
1293 grp->feed = feed; 2462 grp->feed = feed;
1294 2463
1295 grp_try_feed (grp); 2464 grp_try_feed (grp);
1296} 2465}
1297 2466
2467void
1298void eio_grp_limit (eio_req *grp, int limit) 2468eio_grp_limit (eio_req *grp, int limit)
1299{ 2469{
1300 grp->int2 = limit; 2470 grp->int2 = limit;
1301 2471
1302 grp_try_feed (grp); 2472 grp_try_feed (grp);
1303} 2473}
1304 2474
2475void
1305void eio_grp_add (eio_req *grp, eio_req *req) 2476eio_grp_add (eio_req *grp, eio_req *req)
1306{ 2477{
1307 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2478 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2479
2480 grp->flags |= EIO_FLAG_GROUPADD;
1308 2481
1309 ++grp->size; 2482 ++grp->size;
1310 req->grp = grp; 2483 req->grp = grp;
1311 2484
1312 req->grp_prev = 0; 2485 req->grp_prev = 0;
… …
1319} 2492}
1320 2493
1321/*****************************************************************************/ 2494/*****************************************************************************/
1322/* misc garbage */ 2495/* misc garbage */
1323 2496
2497eio_ssize_t
1324ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2498eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325{ 2499{
1326 worker wrk;
1327
1328 wrk.dbuf = 0;
1329
1330 eio__sendfile (ofd, ifd, offset, count, &wrk); 2500 return eio__sendfile (ofd, ifd, offset, count);
1331
1332 if (wrk.dbuf)
1333 free (wrk.dbuf);
1334} 2501}
1335 2502

Diff Legend

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