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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.12 by root, Tue May 13 18:54:52 2008 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 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 X_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#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
68
69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
71#if HAVE_STDINT_H
72# include <stdint.h>
73#endif
74
75#ifndef ECANCELED
76# define ECANCELED EDOM
77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
85
86static void eio_destroy (eio_req *req);
87
15#ifndef EIO_FINISH 88#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 90#endif
18 91
19#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
… …
22 95
23#ifndef EIO_FEED 96#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 98#endif
26 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
27#ifdef _WIN32 111#ifdef _WIN32
28 112
29 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 202
31#else 203#else
32 204
33# include "config.h"
34# include <sys/time.h> 205 #include <sys/time.h>
35# include <sys/select.h> 206 #include <sys/select.h>
207 #include <sys/statvfs.h>
36# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
37# include <utime.h> 243# 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 244#endif
44 245
246#if HAVE_SYS_SYSCALL_H
247# include <sys/syscall.h>
248#endif
249
250#if HAVE_SYS_PRCTL_H
251# include <sys/prctl.h>
45#endif 252#endif
46 253
47#if HAVE_SENDFILE 254#if HAVE_SENDFILE
48# if __linux 255# if __linux
49# include <sys/sendfile.h> 256# include <sys/sendfile.h>
50# elif __freebsd 257# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 258# include <sys/socket.h>
52# include <sys/uio.h> 259# include <sys/uio.h>
53# elif __hpux 260# elif __hpux
54# include <sys/socket.h> 261# include <sys/socket.h>
55# elif __solaris /* not yet */ 262# elif __solaris
56# include <sys/sendfile.h> 263# include <sys/sendfile.h>
57# else 264# else
58# error sendfile support requested but not available 265# error sendfile support requested but not available
59# endif 266# endif
60#endif 267#endif
61 268
62/* number of seconds after which an idle threads exit */ 269#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 270# define D_TYPE(de) 0
271#endif
272#ifndef D_INO
273# define D_INO(de) 0
274#endif
275#ifndef D_NAMLEN
276# define D_NAMLEN(entp) strlen (D_NAME (entp))
277#endif
64 278
65/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 280#ifndef NAME_MAX
67# define NAME_MAX 4096 281# define NAME_MAX 4096
68#endif 282#endif
69 283
284/* used for readlink etc. */
285#ifndef PATH_MAX
286# define PATH_MAX 4096
287#endif
288
70/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
72 291
73#define dBUF \ 292#define dBUF \
74 char *eio_buf; \
75 X_LOCK (wrklock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \
78 errno = ENOMEM; \ 294 errno = ENOMEM; \
79 if (!eio_buf) \ 295 if (!eio_buf) \
80 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
81 300
82#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
302
303/*****************************************************************************/
304
305struct tmpbuf
306{
307 void *ptr;
308 int len;
309};
310
311static void *
312tmpbuf_get (struct tmpbuf *buf, int len)
313{
314 if (buf->len < len)
315 {
316 free (buf->ptr);
317 buf->ptr = malloc (buf->len = len);
318 }
319
320 return buf->ptr;
321}
322
323/*****************************************************************************/
324
325#define ETP_PRI_MIN EIO_PRI_MIN
326#define ETP_PRI_MAX EIO_PRI_MAX
327
328struct etp_worker;
329
330#define ETP_REQ eio_req
331#define ETP_DESTROY(req) eio_destroy (req)
332static int eio_finish (eio_req *req);
333#define ETP_FINISH(req) eio_finish (req)
334static void eio_execute (struct etp_worker *self, eio_req *req);
335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
336
337/*****************************************************************************/
338
339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
340
341/* calculate time difference in ~1/EIO_TICKS of a second */
342ecb_inline int
343tvdiff (struct timeval *tv1, struct timeval *tv2)
344{
345 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
346 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
347}
348
349static unsigned int started, idle, wanted = 4;
83 350
84static void (*want_poll_cb) (void); 351static void (*want_poll_cb) (void);
85static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
86 353
87static unsigned int max_poll_time = 0; 354static unsigned int max_poll_time; /* reslock */
88static unsigned int max_poll_reqs = 0; 355static unsigned int max_poll_reqs; /* reslock */
89 356
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 357static unsigned int nreqs; /* reqlock */
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 358static unsigned int nready; /* reqlock */
92{ 359static unsigned int npending; /* reqlock */
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 360static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 361static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95}
96 362
97static unsigned int started, idle, wanted = 4; 363static xmutex_t wrklock;
364static xmutex_t reslock;
365static xmutex_t reqlock;
366static xcond_t reqwait;
367
368#if !HAVE_PREADWRITE
369/*
370 * make our pread/pwrite emulation safe against themselves, but not against
371 * normal read/write by using a mutex. slows down execution a lot,
372 * but that's your problem, not mine.
373 */
374static xmutex_t preadwritelock;
375#endif
376
377typedef struct etp_worker
378{
379 struct tmpbuf tmpbuf;
380
381 /* locked by wrklock */
382 struct etp_worker *prev, *next;
383
384 xthread_t tid;
385
386#ifdef ETP_WORKER_COMMON
387 ETP_WORKER_COMMON
388#endif
389} etp_worker;
390
391static etp_worker wrk_first; /* NOT etp */
392
393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98 395
99/* worker threads management */ 396/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101 397
102typedef struct worker 398static void ecb_cold
103{
104 /* locked by wrklock */
105 struct worker *prev, *next;
106
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk) 399etp_worker_clear (etp_worker *wrk)
118{ 400{
119 if (wrk->dirp) 401}
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124 402
125 if (wrk->dbuf) 403static void ecb_cold
126 { 404etp_worker_free (etp_worker *wrk)
405{
127 free (wrk->dbuf); 406 free (wrk->tmpbuf.ptr);
128 wrk->dbuf = 0;
129 }
130}
131 407
132static void worker_free (worker *wrk)
133{
134 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
136 410
137 free (wrk); 411 free (wrk);
138} 412}
139 413
140static volatile unsigned int nreqs, nready, npending; 414static unsigned int
141static volatile unsigned int max_idle = 4; 415etp_nreqs (void)
142
143static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT;
146
147unsigned int eio_nreqs (void)
148{ 416{
417 int retval;
418 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
419 retval = nreqs;
420 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149 return nreqs; 421 return retval;
150} 422}
151 423
152unsigned int eio_nready (void) 424static unsigned int
425etp_nready (void)
153{ 426{
154 unsigned int retval; 427 unsigned int retval;
155 428
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 429 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready; 430 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 431 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159 432
160 return retval; 433 return retval;
161} 434}
162 435
163unsigned int eio_npending (void) 436static unsigned int
437etp_npending (void)
164{ 438{
165 unsigned int retval; 439 unsigned int retval;
166 440
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 441 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending; 442 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 443 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170 444
171 return retval; 445 return retval;
172} 446}
173 447
174unsigned int eio_nthreads (void) 448static unsigned int
449etp_nthreads (void)
175{ 450{
176 unsigned int retval; 451 unsigned int retval;
177 452
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 453 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started; 454 retval = started;
… …
186 * a somewhat faster data structure might be nice, but 461 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns 462 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation. 463 * per shift, the most expensive operation.
189 */ 464 */
190typedef struct { 465typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 466 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192 int size; 467 int size;
193} reqq; 468} etp_reqq;
194 469
195static reqq req_queue; 470static etp_reqq req_queue;
196static reqq res_queue; 471static etp_reqq res_queue;
197 472
198static int reqq_push (reqq *q, eio_req *req) 473static void ecb_noinline ecb_cold
474reqq_init (etp_reqq *q)
475{
476 int pri;
477
478 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
479 q->qs[pri] = q->qe[pri] = 0;
480
481 q->size = 0;
482}
483
484static int ecb_noinline
485reqq_push (etp_reqq *q, ETP_REQ *req)
199{ 486{
200 int pri = req->pri; 487 int pri = req->pri;
201 req->next = 0; 488 req->next = 0;
202 489
203 if (q->qe[pri]) 490 if (q->qe[pri])
… …
209 q->qe[pri] = q->qs[pri] = req; 496 q->qe[pri] = q->qs[pri] = req;
210 497
211 return q->size++; 498 return q->size++;
212} 499}
213 500
214static eio_req *reqq_shift (reqq *q) 501static ETP_REQ * ecb_noinline
502reqq_shift (etp_reqq *q)
215{ 503{
216 int pri; 504 int pri;
217 505
218 if (!q->size) 506 if (!q->size)
219 return 0; 507 return 0;
220 508
221 --q->size; 509 --q->size;
222 510
223 for (pri = EIO_NUM_PRI; pri--; ) 511 for (pri = ETP_NUM_PRI; pri--; )
224 { 512 {
225 eio_req *req = q->qs[pri]; 513 eio_req *req = q->qs[pri];
226 514
227 if (req) 515 if (req)
228 { 516 {
… …
234 } 522 }
235 523
236 abort (); 524 abort ();
237} 525}
238 526
239static void grp_try_feed (eio_req *grp) 527static int ecb_cold
528etp_init (void (*want_poll)(void), void (*done_poll)(void))
240{ 529{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 530 X_MUTEX_CREATE (wrklock);
242 { 531 X_MUTEX_CREATE (reslock);
243 int old_len = grp->size; 532 X_MUTEX_CREATE (reqlock);
533 X_COND_CREATE (reqwait);
244 534
245 EIO_FEED (grp); 535 reqq_init (&req_queue);
536 reqq_init (&res_queue);
246 537
247 /* stop if no progress has been made */ 538 wrk_first.next =
248 if (old_len == grp->size) 539 wrk_first.prev = &wrk_first;
249 {
250 grp->feed = 0;
251 break;
252 }
253 }
254}
255 540
256static int eio_finish (eio_req *req); 541 started = 0;
542 idle = 0;
543 nreqs = 0;
544 nready = 0;
545 npending = 0;
257 546
258static int grp_dec (eio_req *grp) 547 want_poll_cb = want_poll;
259{ 548 done_poll_cb = done_poll;
260 --grp->size;
261 549
262 /* call feeder, if applicable */
263 grp_try_feed (grp);
264
265 /* finish, if done */
266 if (!grp->size && grp->int1)
267 return eio_finish (grp);
268 else
269 return 0; 550 return 0;
270} 551}
271 552
272void eio_destroy (eio_req *req)
273{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277 EIO_DESTROY (req);
278}
279
280static int eio_finish (eio_req *req)
281{
282 int res = EIO_FINISH (req);
283
284 if (req->grp)
285 {
286 int res2;
287 eio_req *grp = req->grp;
288
289 /* unlink request */
290 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next;
295
296 res2 = grp_dec (grp);
297
298 if (!res && res2)
299 res = res2;
300 }
301
302 eio_destroy (req);
303
304 return res;
305}
306
307void eio_grp_cancel (eio_req *grp)
308{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next)
310 eio_cancel (grp);
311}
312
313void eio_cancel (eio_req *req)
314{
315 X_LOCK (wrklock);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 553X_THREAD_PROC (etp_proc);
323 554
324static void start_thread (void) 555static void ecb_cold
556etp_start_thread (void)
325{ 557{
326 worker *wrk = calloc (1, sizeof (worker)); 558 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 559
328 /*TODO*/ 560 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk)); 561 assert (("unable to allocate worker thread data", wrk));
330 562
331 X_LOCK (wrklock); 563 X_LOCK (wrklock);
332 564
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 565 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 566 {
335 wrk->prev = &wrk_first; 567 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 568 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 569 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 570 wrk_first.next = wrk;
… …
342 free (wrk); 574 free (wrk);
343 575
344 X_UNLOCK (wrklock); 576 X_UNLOCK (wrklock);
345} 577}
346 578
579static void
347static void maybe_start_thread (void) 580etp_maybe_start_thread (void)
348{ 581{
349 if (eio_nthreads () >= wanted) 582 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 583 return;
351 584
352 /* todo: maybe use idle here, but might be less exact */ 585 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 586 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 587 return;
355 588
356 start_thread (); 589 etp_start_thread ();
357} 590}
358 591
359void eio_submit (eio_req *req) 592static void ecb_cold
593etp_end_thread (void)
360{ 594{
361 req->pri += EIO_PRI_BIAS;
362
363 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366 ++nreqs;
367
368 X_LOCK (reqlock);
369 ++nready;
370 reqq_push (&req_queue, req);
371 X_COND_SIGNAL (reqwait);
372 X_UNLOCK (reqlock);
373
374 maybe_start_thread ();
375}
376
377static void end_thread (void)
378{
379 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
380 596
381 req->type = EIO_QUIT; 597 req->type = -1;
382 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383 599
384 X_LOCK (reqlock); 600 X_LOCK (reqlock);
385 reqq_push (&req_queue, req); 601 reqq_push (&req_queue, req);
386 X_COND_SIGNAL (reqwait); 602 X_COND_SIGNAL (reqwait);
387 X_UNLOCK (reqlock); 603 X_UNLOCK (reqlock);
… …
389 X_LOCK (wrklock); 605 X_LOCK (wrklock);
390 --started; 606 --started;
391 X_UNLOCK (wrklock); 607 X_UNLOCK (wrklock);
392} 608}
393 609
394void eio_set_max_poll_time (double nseconds) 610static int
395{
396 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397 max_poll_time = nseconds;
398 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399}
400
401void eio_set_max_poll_reqs (unsigned int maxreqs)
402{
403 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404 max_poll_reqs = maxreqs;
405 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406}
407
408void eio_set_max_idle (unsigned int nthreads)
409{
410 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411 max_idle = nthreads <= 0 ? 1 : nthreads;
412 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413}
414
415void eio_set_min_parallel (unsigned int nthreads)
416{
417 if (wanted < nthreads)
418 wanted = nthreads;
419}
420
421void eio_set_max_parallel (unsigned int nthreads)
422{
423 if (wanted > nthreads)
424 wanted = nthreads;
425
426 while (started > wanted)
427 end_thread ();
428}
429
430int eio_poll (void) 611etp_poll (void)
431{ 612{
432 int maxreqs = max_poll_reqs; 613 unsigned int maxreqs;
614 unsigned int maxtime;
433 struct timeval tv_start, tv_now; 615 struct timeval tv_start, tv_now;
434 eio_req *req;
435 616
617 X_LOCK (reslock);
618 maxreqs = max_poll_reqs;
619 maxtime = max_poll_time;
620 X_UNLOCK (reslock);
621
436 if (max_poll_time) 622 if (maxtime)
437 gettimeofday (&tv_start, 0); 623 gettimeofday (&tv_start, 0);
438 624
439 for (;;) 625 for (;;)
440 { 626 {
627 ETP_REQ *req;
628
441 maybe_start_thread (); 629 etp_maybe_start_thread ();
442 630
443 X_LOCK (reslock); 631 X_LOCK (reslock);
444 req = reqq_shift (&res_queue); 632 req = reqq_shift (&res_queue);
445 633
446 if (req) 634 if (req)
… …
454 X_UNLOCK (reslock); 642 X_UNLOCK (reslock);
455 643
456 if (!req) 644 if (!req)
457 return 0; 645 return 0;
458 646
647 X_LOCK (reqlock);
459 --nreqs; 648 --nreqs;
649 X_UNLOCK (reqlock);
460 650
461 if (req->type == EIO_GROUP && req->size) 651 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
462 { 652 {
463 req->int1 = 1; /* mark request as delayed */ 653 req->int1 = 1; /* mark request as delayed */
464 continue; 654 continue;
465 } 655 }
466 else 656 else
467 { 657 {
468 int res = eio_finish (req); 658 int res = ETP_FINISH (req);
469 if (res) 659 if (ecb_expect_false (res))
470 return res; 660 return res;
471 } 661 }
472 662
473 if (maxreqs && !--maxreqs) 663 if (ecb_expect_false (maxreqs && !--maxreqs))
474 break; 664 break;
475 665
476 if (max_poll_time) 666 if (maxtime)
477 { 667 {
478 gettimeofday (&tv_now, 0); 668 gettimeofday (&tv_now, 0);
479 669
480 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 670 if (tvdiff (&tv_start, &tv_now) >= maxtime)
481 break; 671 break;
482 } 672 }
483 } 673 }
484 674
485 errno = EAGAIN; 675 errno = EAGAIN;
486 return -1; 676 return -1;
487} 677}
488 678
679static void
680etp_cancel (ETP_REQ *req)
681{
682 req->cancelled = 1;
683
684 eio_grp_cancel (req);
685}
686
687static void
688etp_submit (ETP_REQ *req)
689{
690 req->pri -= ETP_PRI_MIN;
691
692 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
693 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
694
695 if (ecb_expect_false (req->type == EIO_GROUP))
696 {
697 /* I hope this is worth it :/ */
698 X_LOCK (reqlock);
699 ++nreqs;
700 X_UNLOCK (reqlock);
701
702 X_LOCK (reslock);
703
704 ++npending;
705
706 if (!reqq_push (&res_queue, req) && want_poll_cb)
707 want_poll_cb ();
708
709 X_UNLOCK (reslock);
710 }
711 else
712 {
713 X_LOCK (reqlock);
714 ++nreqs;
715 ++nready;
716 reqq_push (&req_queue, req);
717 X_COND_SIGNAL (reqwait);
718 X_UNLOCK (reqlock);
719
720 etp_maybe_start_thread ();
721 }
722}
723
724static void ecb_cold
725etp_set_max_poll_time (double nseconds)
726{
727 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
728 max_poll_time = nseconds * EIO_TICKS;
729 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
730}
731
732static void ecb_cold
733etp_set_max_poll_reqs (unsigned int maxreqs)
734{
735 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
736 max_poll_reqs = maxreqs;
737 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
738}
739
740static void ecb_cold
741etp_set_max_idle (unsigned int nthreads)
742{
743 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
744 max_idle = nthreads;
745 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
746}
747
748static void ecb_cold
749etp_set_idle_timeout (unsigned int seconds)
750{
751 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
752 idle_timeout = seconds;
753 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
754}
755
756static void ecb_cold
757etp_set_min_parallel (unsigned int nthreads)
758{
759 if (wanted < nthreads)
760 wanted = nthreads;
761}
762
763static void ecb_cold
764etp_set_max_parallel (unsigned int nthreads)
765{
766 if (wanted > nthreads)
767 wanted = nthreads;
768
769 while (started > wanted)
770 etp_end_thread ();
771}
772
773/*****************************************************************************/
774
775static void
776grp_try_feed (eio_req *grp)
777{
778 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
779 {
780 grp->flags &= ~EIO_FLAG_GROUPADD;
781
782 EIO_FEED (grp);
783
784 /* stop if no progress has been made */
785 if (!(grp->flags & EIO_FLAG_GROUPADD))
786 {
787 grp->feed = 0;
788 break;
789 }
790 }
791}
792
793static int
794grp_dec (eio_req *grp)
795{
796 --grp->size;
797
798 /* call feeder, if applicable */
799 grp_try_feed (grp);
800
801 /* finish, if done */
802 if (!grp->size && grp->int1)
803 return eio_finish (grp);
804 else
805 return 0;
806}
807
808static void
809eio_destroy (eio_req *req)
810{
811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
813
814 EIO_DESTROY (req);
815}
816
817static int
818eio_finish (eio_req *req)
819{
820 int res = EIO_FINISH (req);
821
822 if (req->grp)
823 {
824 int res2;
825 eio_req *grp = req->grp;
826
827 /* unlink request */
828 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
829 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
830
831 if (grp->grp_first == req)
832 grp->grp_first = req->grp_next;
833
834 res2 = grp_dec (grp);
835
836 if (!res)
837 res = res2;
838 }
839
840 eio_destroy (req);
841
842 return res;
843}
844
845void
846eio_grp_cancel (eio_req *grp)
847{
848 for (grp = grp->grp_first; grp; grp = grp->grp_next)
849 eio_cancel (grp);
850}
851
852void
853eio_cancel (eio_req *req)
854{
855 etp_cancel (req);
856}
857
858void
859eio_submit (eio_req *req)
860{
861 etp_submit (req);
862}
863
864unsigned int
865eio_nreqs (void)
866{
867 return etp_nreqs ();
868}
869
870unsigned int
871eio_nready (void)
872{
873 return etp_nready ();
874}
875
876unsigned int
877eio_npending (void)
878{
879 return etp_npending ();
880}
881
882unsigned int ecb_cold
883eio_nthreads (void)
884{
885 return etp_nthreads ();
886}
887
888void ecb_cold
889eio_set_max_poll_time (double nseconds)
890{
891 etp_set_max_poll_time (nseconds);
892}
893
894void ecb_cold
895eio_set_max_poll_reqs (unsigned int maxreqs)
896{
897 etp_set_max_poll_reqs (maxreqs);
898}
899
900void ecb_cold
901eio_set_max_idle (unsigned int nthreads)
902{
903 etp_set_max_idle (nthreads);
904}
905
906void ecb_cold
907eio_set_idle_timeout (unsigned int seconds)
908{
909 etp_set_idle_timeout (seconds);
910}
911
912void ecb_cold
913eio_set_min_parallel (unsigned int nthreads)
914{
915 etp_set_min_parallel (nthreads);
916}
917
918void ecb_cold
919eio_set_max_parallel (unsigned int nthreads)
920{
921 etp_set_max_parallel (nthreads);
922}
923
924int eio_poll (void)
925{
926 return etp_poll ();
927}
928
489/*****************************************************************************/ 929/*****************************************************************************/
490/* work around various missing functions */ 930/* work around various missing functions */
491 931
492#if !HAVE_PREADWRITE 932#if !HAVE_PREADWRITE
933# undef pread
934# undef pwrite
493# define pread eio__pread 935# define pread eio__pread
494# define pwrite eio__pwrite 936# define pwrite eio__pwrite
495 937
496/*
497 * make our pread/pwrite safe against themselves, but not against
498 * normal read/write by using a mutex. slows down execution a lot,
499 * but that's your problem, not mine.
500 */
501static mutex_t preadwritelock = X_MUTEX_INIT;
502
503static ssize_t 938static eio_ssize_t
504eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
505{ 940{
506 ssize_t res; 941 eio_ssize_t res;
507 off_t ooffset; 942 off_t ooffset;
508 943
509 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
510 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
511 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
… …
514 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
515 950
516 return res; 951 return res;
517} 952}
518 953
519static ssize_t 954static eio_ssize_t
520eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521{ 956{
522 ssize_t res; 957 eio_ssize_t res;
523 off_t ooffset; 958 off_t ooffset;
524 959
525 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
526 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
527 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
528 res = write (fd, buf, count); 963 res = write (fd, buf, count);
529 lseek (fd, offset, SEEK_SET); 964 lseek (fd, ooffset, SEEK_SET);
530 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
531 966
532 return res; 967 return res;
533} 968}
534#endif 969#endif
535 970
536#ifndef HAVE_FUTIMES 971#ifndef HAVE_UTIMES
537 972
973# undef utimes
538# define utimes(path,times) eio__utimes (path, times) 974# define utimes(path,times) eio__utimes (path, times)
539# define futimes(fd,times) eio__futimes (fd, times)
540 975
541static int 976static int
542eio__utimes (const char *filename, const struct timeval times[2]) 977eio__utimes (const char *filename, const struct timeval times[2])
543{ 978{
544 if (times) 979 if (times)
… …
552 } 987 }
553 else 988 else
554 return utime (filename, 0); 989 return utime (filename, 0);
555} 990}
556 991
992#endif
993
994#ifndef HAVE_FUTIMES
995
996# undef futimes
997# define futimes(fd,times) eio__futimes (fd, times)
998
999static int
557static int eio__futimes (int fd, const struct timeval tv[2]) 1000eio__futimes (int fd, const struct timeval tv[2])
558{ 1001{
559 errno = ENOSYS; 1002 errno = ENOSYS;
560 return -1; 1003 return -1;
561} 1004}
562 1005
563#endif 1006#endif
564 1007
565#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
1009# undef fdatasync
566# define fdatasync fsync 1010# define fdatasync(fd) fsync (fd)
567#endif 1011#endif
1012
1013static int
1014eio__syncfs (int fd)
1015{
1016 int res;
1017
1018#if HAVE_SYS_SYNCFS
1019 res = (int)syscall (__NR_syncfs, (int)(fd));
1020#else
1021 res = -1;
1022 errno = ENOSYS;
1023#endif
1024
1025 if (res < 0 && errno == ENOSYS && fd >= 0)
1026 sync ();
1027
1028 return res;
1029}
1030
1031/* sync_file_range always needs emulation */
1032static int
1033eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1034{
1035#if HAVE_SYNC_FILE_RANGE
1036 int res;
1037
1038 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1039 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1040 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1041 {
1042 flags = 0
1043 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1044 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1045 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1046 }
1047
1048 res = sync_file_range (fd, offset, nbytes, flags);
1049
1050 if (!res || errno != ENOSYS)
1051 return res;
1052#endif
1053
1054 /* even though we could play tricks with the flags, it's better to always
1055 * call fdatasync, as that matches the expectation of its users best */
1056 return fdatasync (fd);
1057}
1058
1059static int
1060eio__fallocate (int fd, int mode, off_t offset, size_t len)
1061{
1062#if HAVE_FALLOCATE
1063 return fallocate (fd, mode, offset, len);
1064#else
1065 errno = ENOSYS;
1066 return -1;
1067#endif
1068}
568 1069
569#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
1071# undef readahead
570# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571 1073
572static ssize_t 1074static eio_ssize_t
573eio__readahead (int fd, off_t offset, size_t count, worker *self) 1075eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574{ 1076{
575 size_t todo = count; 1077 size_t todo = count;
576 dBUF; 1078 dBUF;
577 1079
578 while (todo > 0) 1080 while (todo > 0)
… …
582 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
583 offset += len; 1085 offset += len;
584 todo -= len; 1086 todo -= len;
585 } 1087 }
586 1088
1089 FUBd;
1090
587 errno = 0; 1091 errno = 0;
588 return count; 1092 return count;
589} 1093}
590 1094
591#endif 1095#endif
592 1096
593#if !HAVE_READDIR_R
594# define readdir_r eio__readdir_r
595
596static mutex_t readdirlock = X_MUTEX_INIT;
597
598static int
599eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600{
601 EIO_STRUCT_DIRENT *e;
602 int errorno;
603
604 X_LOCK (readdirlock);
605
606 e = readdir (dirp);
607 errorno = errno;
608
609 if (e)
610 {
611 *res = ent;
612 strcpy (ent->d_name, e->d_name);
613 }
614 else
615 *res = 0;
616
617 X_UNLOCK (readdirlock);
618
619 errno = errorno;
620 return e ? 0 : -1;
621}
622#endif
623
624/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
625static ssize_t 1098static eio_ssize_t
626eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 1099eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627{ 1100{
1101 eio_ssize_t written = 0;
628 ssize_t res; 1102 eio_ssize_t res;
629 1103
630 if (!count) 1104 if (!count)
631 return 0; 1105 return 0;
632 1106
1107 for (;;)
1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
633#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
634# if __linux 1113# if __linux
1114 off_t soffset = offset;
635 res = sendfile (ofd, ifd, &offset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
636 1116
637# elif __freebsd 1117# elif __FreeBSD__
638 /* 1118 /*
639 * Of course, the freebsd sendfile is a dire hack with no thoughts 1119 * Of course, the freebsd sendfile is a dire hack with no thoughts
640 * wasted on making it similar to other I/O functions. 1120 * wasted on making it similar to other I/O functions.
641 */ 1121 */
642 {
643 off_t sbytes; 1122 off_t sbytes;
644 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1123 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645 1124
646 if (res < 0 && sbytes) 1125 #if 0 /* according to the manpage, this is correct, but broken behaviour */
647 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1126 /* freebsd' sendfile will return 0 on success */
1127 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1128 /* not on e.g. EIO or EPIPE - sounds broken */
1129 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648 res = sbytes; 1130 res = sbytes;
649 } 1131 #endif
1132
1133 /* according to source inspection, this is correct, and useful behaviour */
1134 if (sbytes)
1135 res = sbytes;
1136
1137# elif defined (__APPLE__)
1138 off_t sbytes = count;
1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1140
1141 /* according to the manpage, sbytes is always valid */
1142 if (sbytes)
1143 res = sbytes;
650 1144
651# elif __hpux 1145# elif __hpux
652 res = sendfile (ofd, ifd, offset, count, 0, 0); 1146 res = sendfile (ofd, ifd, offset, count, 0, 0);
653 1147
654# elif __solaris 1148# elif __solaris
655 {
656 struct sendfilevec vec; 1149 struct sendfilevec vec;
657 size_t sbytes; 1150 size_t sbytes;
658 1151
659 vec.sfv_fd = ifd; 1152 vec.sfv_fd = ifd;
660 vec.sfv_flag = 0; 1153 vec.sfv_flag = 0;
661 vec.sfv_off = offset; 1154 vec.sfv_off = offset;
662 vec.sfv_len = count; 1155 vec.sfv_len = count;
663 1156
664 res = sendfilev (ofd, &vec, 1, &sbytes); 1157 res = sendfilev (ofd, &vec, 1, &sbytes);
665 1158
666 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
667 res = sbytes; 1160 res = sbytes;
668 }
669 1161
670# endif 1162# endif
1163
1164#elif defined (_WIN32) && 0
1165 /* does not work, just for documentation of what would need to be done */
1166 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1167 /* libeio guarantees that the file offset does not change, and windows */
1168 /* has no way to get an independent handle to the same file description */
1169 HANDLE h = TO_SOCKET (ifd);
1170 SetFilePointer (h, offset, 0, FILE_BEGIN);
1171 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1172
671#else 1173#else
672 res = -1; 1174 res = -1;
673 errno = ENOSYS; 1175 errno = ENOSYS;
674#endif 1176#endif
675 1177
1178 /* we assume sendfile can copy at least 128mb in one go */
1179 if (res <= 128 * 1024 * 1024)
1180 {
1181 if (res > 0)
1182 written += res;
1183
1184 if (written)
1185 return written;
1186
1187 break;
1188 }
1189 else
1190 {
1191 /* if we requested more, then probably the kernel was lazy */
1192 written += res;
1193 offset += res;
1194 count -= res;
1195
1196 if (!count)
1197 return written;
1198 }
1199 }
1200
676 if (res < 0 1201 if (res < 0
677 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1203 /* BSDs */
1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1205 || errno == ENOTSUP
1206#endif
1207#ifdef EOPNOTSUPP /* windows */
1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
678#if __solaris 1210#if __solaris
679 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
680#endif 1212#endif
681 ) 1213 )
682 ) 1214 )
… …
686 1218
687 res = 0; 1219 res = 0;
688 1220
689 while (count) 1221 while (count)
690 { 1222 {
691 ssize_t cnt; 1223 eio_ssize_t cnt;
692 1224
693 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1225 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694 1226
695 if (cnt <= 0) 1227 if (cnt <= 0)
696 { 1228 {
… …
708 1240
709 offset += cnt; 1241 offset += cnt;
710 res += cnt; 1242 res += cnt;
711 count -= cnt; 1243 count -= cnt;
712 } 1244 }
1245
1246 FUBd;
713 } 1247 }
714 1248
715 return res; 1249 return res;
1250}
1251
1252#ifdef PAGESIZE
1253# define eio_pagesize() PAGESIZE
1254#else
1255static intptr_t
1256eio_pagesize (void)
1257{
1258 static intptr_t page;
1259
1260 if (!page)
1261 page = sysconf (_SC_PAGESIZE);
1262
1263 return page;
1264}
1265#endif
1266
1267static void
1268eio_page_align (void **addr, size_t *length)
1269{
1270 intptr_t mask = eio_pagesize () - 1;
1271
1272 /* round down addr */
1273 intptr_t adj = mask & (intptr_t)*addr;
1274
1275 *addr = (void *)((intptr_t)*addr - adj);
1276 *length += adj;
1277
1278 /* round up length */
1279 *length = (*length + mask) & ~mask;
1280}
1281
1282#if !_POSIX_MEMLOCK
1283# define eio__mlockall(a) EIO_ENOSYS ()
1284#else
1285
1286static int
1287eio__mlockall (int flags)
1288{
1289 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1290 extern int mallopt (int, int);
1291 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1292 #endif
1293
1294 if (EIO_MCL_CURRENT != MCL_CURRENT
1295 || EIO_MCL_FUTURE != MCL_FUTURE)
1296 {
1297 flags = 0
1298 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1299 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1300 }
1301
1302 return mlockall (flags);
1303}
1304#endif
1305
1306#if !_POSIX_MEMLOCK_RANGE
1307# define eio__mlock(a,b) EIO_ENOSYS ()
1308#else
1309
1310static int
1311eio__mlock (void *addr, size_t length)
1312{
1313 eio_page_align (&addr, &length);
1314
1315 return mlock (addr, length);
1316}
1317
1318#endif
1319
1320#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1321# define eio__msync(a,b,c) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__msync (void *mem, size_t len, int flags)
1326{
1327 eio_page_align (&mem, &len);
1328
1329 if (EIO_MS_ASYNC != MS_SYNC
1330 || EIO_MS_INVALIDATE != MS_INVALIDATE
1331 || EIO_MS_SYNC != MS_SYNC)
1332 {
1333 flags = 0
1334 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1335 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1336 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1337 }
1338
1339 return msync (mem, len, flags);
1340}
1341
1342#endif
1343
1344static int
1345eio__mtouch (eio_req *req)
1346{
1347 void *mem = req->ptr2;
1348 size_t len = req->size;
1349 int flags = req->int1;
1350
1351 eio_page_align (&mem, &len);
1352
1353 {
1354 intptr_t addr = (intptr_t)mem;
1355 intptr_t end = addr + len;
1356 intptr_t page = eio_pagesize ();
1357
1358 if (addr < end)
1359 if (flags & EIO_MT_MODIFY) /* modify */
1360 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1361 else
1362 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1363 }
1364
1365 return 0;
1366}
1367
1368/*****************************************************************************/
1369/* requests implemented outside eio_execute, because they are so large */
1370
1371/* copies some absolute path to tmpbuf */
1372static char *
1373eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1374{
1375 if (wd == EIO_CWD)
1376 return getcwd (tmpbuf->ptr, PATH_MAX);
1377
1378#if HAVE_AT
1379 abort (); /*TODO*/
1380#else
1381 strcpy (tmpbuf->ptr, wd);
1382#endif
1383 return tmpbuf->ptr;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 /*D*/ /*TODO: wd ignored */
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 if (!eio__getwd (tmpbuf, wd))
1439 return -1;
1440
1441 if (res [1]) /* only use if not / */
1442 res += strlen (res);
1443 }
1444
1445 while (*rel)
1446 {
1447 eio_ssize_t len, linklen;
1448 const char *beg = rel;
1449
1450 while (*rel && *rel != '/')
1451 ++rel;
1452
1453 len = rel - beg;
1454
1455 if (!len) /* skip slashes */
1456 {
1457 ++rel;
1458 continue;
1459 }
1460
1461 if (beg [0] == '.')
1462 {
1463 if (len == 1)
1464 continue; /* . - nop */
1465
1466 if (beg [1] == '.' && len == 2)
1467 {
1468 /* .. - back up one component, if possible */
1469
1470 while (res != tmpbuf->ptr)
1471 if (*--res == '/')
1472 break;
1473
1474 continue;
1475 }
1476 }
1477
1478 errno = ENAMETOOLONG;
1479 if (res + 1 + len + 1 >= tmp1)
1480 return;
1481
1482 /* copy one component */
1483 *res = '/';
1484 memcpy (res + 1, beg, len);
1485
1486 /* zero-terminate, for readlink */
1487 res [len + 1] = 0;
1488
1489 /* now check if it's a symlink */
1490 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1491
1492 if (linklen < 0)
1493 {
1494 if (errno != EINVAL)
1495 return -1;
1496
1497 /* it's a normal directory. hopefully */
1498 res += len + 1;
1499 }
1500 else
1501 {
1502 /* yay, it was a symlink - build new path in tmp2 */
1503 int rellen = strlen (rel);
1504
1505 errno = ENAMETOOLONG;
1506 if (linklen + 1 + rellen >= PATH_MAX)
1507 return -1;
1508
1509 errno = ELOOP;
1510 if (!--symlinks)
1511 return -1;
1512
1513 if (*tmp1 == '/')
1514 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1515
1516 /* we need to be careful, as rel might point into tmp2 already */
1517 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1518 tmp2 [linklen] = '/';
1519 memcpy (tmp2, tmp1, linklen);
1520
1521 rel = tmp2;
1522 }
1523 }
1524
1525 /* special case for the lone root path */
1526 if (res == tmpbuf->ptr)
1527 *res++ = '/';
1528
1529 return res - (char *)tmpbuf->ptr;
1530}
1531
1532static signed char
1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1534{
1535 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1536 : a->inode < b->inode ? -1
1537 : a->inode > b->inode ? 1
1538 : 0;
1539}
1540
1541#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1542
1543#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1545
1546static void
1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1548{
1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1550 unsigned char *bit = bits;
1551
1552 assert (CHAR_BIT == 8);
1553 assert (sizeof (eio_dirent) * 8 < 256);
1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1555 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1556
1557 if (size <= EIO_SORT_FAST)
1558 return;
1559
1560 /* first prepare an array of bits to test in our radix sort */
1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1562 /* inode_bits must contain all inodes ORed together */
1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1564 {
1565 eio_ino_t endianness;
1566 int i, j;
1567
1568 /* we store the byte offset of byte n into byte n of "endianness" */
1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1570 ((unsigned char *)&endianness)[i] = i;
1571
1572 *bit++ = 0;
1573
1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1575 {
1576 /* shifting off the byte offsets out of "endianness" */
1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1578 endianness >>= 8;
1579
1580 for (j = 0; j < 8; ++j)
1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1582 *bit++ = offs + j;
1583 }
1584
1585 for (j = 0; j < 8; ++j)
1586 if (score_bits & (1 << j))
1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1588 }
1589
1590 /* now actually do the sorting (a variant of MSD radix sort) */
1591 {
1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1595 int stk_idx = 0;
1596
1597 base_stk [stk_idx] = dents;
1598 end_stk [stk_idx] = dents + size;
1599 bit_stk [stk_idx] = bit - 1;
1600
1601 do
1602 {
1603 base = base_stk [stk_idx];
1604 end = end_stk [stk_idx];
1605 bit = bit_stk [stk_idx];
1606
1607 for (;;)
1608 {
1609 unsigned char O = *bit >> 3;
1610 unsigned char M = 1 << (*bit & 7);
1611
1612 eio_dirent *a = base;
1613 eio_dirent *b = end;
1614
1615 if (b - a < EIO_SORT_CUTOFF)
1616 break;
1617
1618 /* now bit-partition the array on the bit */
1619 /* this ugly asymmetric loop seems to perform much better than typical */
1620 /* partition algos found in the literature */
1621 do
1622 if (!(((unsigned char *)a)[O] & M))
1623 ++a;
1624 else if (!(((unsigned char *)--b)[O] & M))
1625 {
1626 eio_dirent tmp = *a; *a = *b; *b = tmp;
1627 ++a;
1628 }
1629 while (b > a);
1630
1631 /* next bit, or stop, if no bits left in this path */
1632 if (!*--bit)
1633 break;
1634
1635 base_stk [stk_idx] = a;
1636 end_stk [stk_idx] = end;
1637 bit_stk [stk_idx] = bit;
1638 ++stk_idx;
1639
1640 end = a;
1641 }
1642 }
1643 while (stk_idx--);
1644 }
1645}
1646
1647static void
1648eio_dent_insertion_sort (eio_dirent *dents, int size)
1649{
1650 /* first move the smallest element to the front, to act as a sentinel */
1651 {
1652 int i;
1653 eio_dirent *min = dents;
1654
1655 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1656 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1657 if (EIO_DENT_CMP (dents [i], <, *min))
1658 min = &dents [i];
1659
1660 /* swap elements 0 and j (minimum) */
1661 {
1662 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1663 }
1664 }
1665
1666 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1667 {
1668 eio_dirent *i, *j;
1669
1670 for (i = dents + 1; i < dents + size; ++i)
1671 {
1672 eio_dirent value = *i;
1673
1674 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1675 j [1] = j [0];
1676
1677 j [1] = value;
1678 }
1679 }
1680}
1681
1682static void
1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1684{
1685 if (size <= 1)
1686 return; /* our insertion sort relies on size > 0 */
1687
1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1689 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1690 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1691
1692 /* use an insertion sort at the end, or for small arrays, */
1693 /* as insertion sort is more efficient for small partitions */
1694 eio_dent_insertion_sort (dents, size);
716} 1695}
717 1696
718/* read a full directory */ 1697/* read a full directory */
719static void 1698static void
720eio__scandir (eio_req *req, worker *self) 1699eio__scandir (eio_req *req)
721{ 1700{
1701 char *name, *names;
1702 int namesalloc = 4096 - sizeof (void *) * 4;
1703 int namesoffs = 0;
1704 int flags = req->int1;
1705 eio_dirent *dents = 0;
1706 int dentalloc = 128;
1707 int dentoffs = 0;
1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
722 DIR *dirp; 1713 DIR *dirp;
723 union
724 {
725 EIO_STRUCT_DIRENT d;
726 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727 } *u;
728 EIO_STRUCT_DIRENT *entp; 1714 EIO_STRUCT_DIRENT *entp;
729 char *name, *names; 1715#endif
730 int memlen = 4096;
731 int memofs = 0;
732 int res = 0;
733 1716
734 X_LOCK (wrklock); 1717 req->result = -1;
735 self->dirp = dirp = opendir (req->ptr1);
736 self->dbuf = u = malloc (sizeof (*u));
737 req->flags |= EIO_FLAG_PTR2_FREE;
738 req->ptr2 = names = malloc (memlen);
739 X_UNLOCK (wrklock);
740 1718
741 if (dirp && u && names) 1719 if (!(flags & EIO_READDIR_DENTS))
742 for (;;) 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1721
1722#ifdef _WIN32
1723 {
1724 int len = strlen ((const char *)req->ptr1);
1725 char *path = malloc (MAX_PATH);
1726 const char *fmt;
1727
1728 if (!len)
1729 fmt = "./*";
1730 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1731 fmt = "%s*";
1732 else
1733 fmt = "%s/*";
1734
1735 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1736 dirp = FindFirstFile (path, &entp);
1737 free (path);
1738
1739 if (dirp == INVALID_HANDLE_VALUE)
743 { 1740 {
1741 /* should steal _dosmaperr */
1742 switch (GetLastError ())
1743 {
1744 case ERROR_FILE_NOT_FOUND:
1745 req->result = 0;
1746 break;
1747
1748 case ERROR_INVALID_NAME:
1749 case ERROR_PATH_NOT_FOUND:
1750 case ERROR_NO_MORE_FILES:
1751 errno = ENOENT;
1752 break;
1753
1754 case ERROR_NOT_ENOUGH_MEMORY:
1755 errno = ENOMEM;
1756 break;
1757
1758 default:
1759 errno = EINVAL;
1760 break;
1761 }
1762
1763 return;
1764 }
1765 }
1766#else
1767 dirp = opendir (req->ptr1);
1768
1769 if (!dirp)
1770 return;
1771#endif
1772
1773 if (req->flags & EIO_FLAG_PTR1_FREE)
1774 free (req->ptr1);
1775
1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1778 req->ptr2 = names = malloc (namesalloc);
1779
1780 if (!names || (flags && !dents))
1781 return;
1782
1783 for (;;)
1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
744 errno = 0; 1790 errno = 0;
745 readdir_r (dirp, &u->d, &entp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
746 1794
1795 if (done)
1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
747 if (!entp) 1802 if (errno)
1803 break;
1804#endif
1805
1806 /* sort etc. */
1807 req->int1 = flags;
1808 req->result = dentoffs;
1809
1810 if (flags & EIO_READDIR_STAT_ORDER)
1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1815 else
1816 {
1817 /* in this case, all is known, and we just put dirs first and sort them */
1818 eio_dirent *oth = dents + dentoffs;
1819 eio_dirent *dir = dents;
1820
1821 /* now partition dirs to the front, and non-dirs to the back */
1822 /* by walking from both sides and swapping if necessary */
1823 while (oth > dir)
1824 {
1825 if (dir->type == EIO_DT_DIR)
1826 ++dir;
1827 else if ((--oth)->type == EIO_DT_DIR)
1828 {
1829 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1830
1831 ++dir;
1832 }
1833 }
1834
1835 /* now sort the dirs only (dirs all have the same score) */
1836 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1837 }
1838
748 break; 1839 break;
1840 }
749 1841
750 name = entp->d_name; 1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
751 1844
1845 /* skip . and .. entries */
752 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753 { 1847 {
754 int len = strlen (name) + 1; 1848 int len = D_NAMLEN (entp) + 1;
755 1849
756 res++; 1850 while (ecb_expect_false (namesoffs + len > namesalloc))
757
758 while (memofs + len > memlen)
759 { 1851 {
760 memlen *= 2; 1852 namesalloc *= 2;
761 X_LOCK (wrklock);
762 req->ptr2 = names = realloc (names, memlen); 1853 req->ptr2 = names = realloc (names, namesalloc);
763 X_UNLOCK (wrklock);
764 1854
765 if (!names) 1855 if (!names)
1856 break;
1857 }
1858
1859 memcpy (names + namesoffs, name, len);
1860
1861 if (dents)
1862 {
1863 struct eio_dirent *ent;
1864
1865 if (ecb_expect_false (dentoffs == dentalloc))
1866 {
1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
766 break; 1871 break;
767 } 1872 }
768 1873
769 memcpy (names + memofs, name, len); 1874 ent = dents + dentoffs;
770 memofs += len; 1875
1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1877 ent->namelen = len - 1;
1878 ent->inode = D_INO (entp);
1879
1880 inode_bits |= ent->inode;
1881
1882 switch (D_TYPE (entp))
1883 {
1884 default:
1885 ent->type = EIO_DT_UNKNOWN;
1886 flags |= EIO_READDIR_FOUND_UNKNOWN;
1887 break;
1888
1889 #ifdef DT_FIFO
1890 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1891 #endif
1892 #ifdef DT_CHR
1893 case DT_CHR: ent->type = EIO_DT_CHR; break;
1894 #endif
1895 #ifdef DT_MPC
1896 case DT_MPC: ent->type = EIO_DT_MPC; break;
1897 #endif
1898 #ifdef DT_DIR
1899 case DT_DIR: ent->type = EIO_DT_DIR; break;
1900 #endif
1901 #ifdef DT_NAM
1902 case DT_NAM: ent->type = EIO_DT_NAM; break;
1903 #endif
1904 #ifdef DT_BLK
1905 case DT_BLK: ent->type = EIO_DT_BLK; break;
1906 #endif
1907 #ifdef DT_MPB
1908 case DT_MPB: ent->type = EIO_DT_MPB; break;
1909 #endif
1910 #ifdef DT_REG
1911 case DT_REG: ent->type = EIO_DT_REG; break;
1912 #endif
1913 #ifdef DT_NWK
1914 case DT_NWK: ent->type = EIO_DT_NWK; break;
1915 #endif
1916 #ifdef DT_CMP
1917 case DT_CMP: ent->type = EIO_DT_CMP; break;
1918 #endif
1919 #ifdef DT_LNK
1920 case DT_LNK: ent->type = EIO_DT_LNK; break;
1921 #endif
1922 #ifdef DT_SOCK
1923 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1924 #endif
1925 #ifdef DT_DOOR
1926 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1927 #endif
1928 #ifdef DT_WHT
1929 case DT_WHT: ent->type = EIO_DT_WHT; break;
1930 #endif
1931 }
1932
1933 ent->score = 7;
1934
1935 if (flags & EIO_READDIR_DIRS_FIRST)
1936 {
1937 if (ent->type == EIO_DT_UNKNOWN)
1938 {
1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1940 ent->score = 1;
1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1943 }
1944 else if (ent->type == EIO_DT_DIR)
1945 ent->score = 0;
1946 }
771 } 1947 }
1948
1949 namesoffs += len;
1950 ++dentoffs;
772 } 1951 }
773 1952
774 if (errno) 1953 if (EIO_CANCELLED (req))
775 res = -1; 1954 {
1955 errno = ECANCELED;
1956 break;
1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1964 }
1965#endif
1966 }
1967}
1968
1969/*****************************************************************************/
1970/* working directory stuff */
1971
1972#if HAVE_AT
1973
1974#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1975
1976#ifndef O_SEARCH
1977# define O_SEARCH O_RDONLY
1978#endif
1979
1980eio_wd
1981eio_wd_open_sync (eio_wd wd, const char *path)
1982{
1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1984
1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1986}
1987
1988static eio_wd
1989eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1990{
1991 return eio_wd_open_sync (wd, path);
1992}
1993
1994void
1995eio_wd_close_sync (eio_wd wd)
1996{
1997 int fd = WD2FD (wd);
1998
1999 if (fd >= 0)
2000 close (fd);
2001}
2002
2003static int
2004eio__truncateat (int dirfd, const char *path, off_t length)
2005{
2006 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2007 int res;
2008
2009 if (fd < 0)
2010 return fd;
2011
2012 res = ftruncate (fd, length);
2013 close (fd);
2014 return res;
2015}
2016
2017static int
2018eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2019{
2020 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2021 int res;
2022
2023 if (fd < 0)
2024 return fd;
2025
2026 res = fstatvfs (fd, buf);
2027 close (fd);
2028 return res;
2029
2030}
2031
2032#else
2033
2034/* on legacy systems, we represent the working directories simply by their path strings */
2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2038{
2039 if (!wd || *path == '/')
2040 return path;
2041
776 2042 {
777 req->result = res; 2043 int l1 = strlen ((const char *)wd);
2044 int l2 = strlen (path);
2045
2046 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2047
2048 memcpy (res, wd, l1);
2049 res [l1] = '/';
2050 memcpy (res + l1 + 1, path, l2 + 1);
2051
2052 return res;
2053 }
778} 2054}
2055
2056static eio_wd
2057eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2058{
2059 if (*path == '/') /* absolute paths ignore wd */
2060 path = strdup (path);
2061 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2062 return wd;
2063 else
2064 {
2065 int len = eio__realpath (tmpbuf, wd, path);
2066
2067 path = EIO_INVALID_WD;
2068
2069 if (len >= 0)
2070 {
2071 ((char *)tmpbuf->ptr)[len] = 0;
2072 path = strdup (tmpbuf->ptr);
2073 }
2074 }
2075
2076 if (!path)
2077 path = EIO_INVALID_WD;
2078
2079 return (eio_wd)path;
2080}
2081
2082eio_wd
2083eio_wd_open_sync (eio_wd wd, const char *path)
2084{
2085 struct tmpbuf tmpbuf = { 0 };
2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
2088
2089 return wd;
2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
779 2100
780/*****************************************************************************/ 2101/*****************************************************************************/
781 2102
782#define ALLOC(len) \ 2103#define ALLOC(len) \
783 if (!req->ptr2) \ 2104 if (!req->ptr2) \
… …
792 req->result = -1; \ 2113 req->result = -1; \
793 break; \ 2114 break; \
794 } \ 2115 } \
795 } 2116 }
796 2117
797X_THREAD_PROC (eio_proc) 2118X_THREAD_PROC (etp_proc)
798{ 2119{
799 eio_req *req; 2120 ETP_REQ *req;
800 struct timespec ts; 2121 struct timespec ts;
801 worker *self = (worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
802 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
803 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
804 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805 2130
806 for (;;) 2131 for (;;)
807 { 2132 {
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2133 ts.tv_sec = 0;
809 2134
810 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
811 2136
812 for (;;) 2137 for (;;)
813 { 2138 {
814 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
815 2140
816 if (req) 2141 if (req)
817 break; 2142 break;
818 2143
2144 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2145 {
2146 X_UNLOCK (reqlock);
2147 X_LOCK (wrklock);
2148 --started;
2149 X_UNLOCK (wrklock);
2150 goto quit;
2151 }
2152
819 ++idle; 2153 ++idle;
820 2154
821 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2155 if (idle <= max_idle)
2156 /* we are allowed to idle, so do so without any timeout */
2157 X_COND_WAIT (reqwait, reqlock);
2158 else
822 { 2159 {
823 if (idle > max_idle) 2160 /* initialise timeout once */
824 { 2161 if (!ts.tv_sec)
825 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
826 X_UNLOCK (reqlock);
827 X_LOCK (wrklock);
828 --started;
829 X_UNLOCK (wrklock);
830 goto quit;
831 }
832 2163
833 /* we are allowed to idle, so do so without any timeout */
834 X_COND_WAIT (reqwait, reqlock); 2164 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
835 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2165 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
836 } 2166 }
837 2167
838 --idle; 2168 --idle;
839 } 2169 }
840 2170
841 --nready; 2171 --nready;
842 2172
843 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
844 2174
845 errno = 0; /* strictly unnecessary */
846
847 if (!EIO_CANCELLED (req))
848 switch (req->type) 2175 if (req->type < 0)
849 {
850 case EIO_READ: ALLOC (req->size);
851 req->result = req->offs >= 0
852 ? pread (req->int1, req->ptr2, req->size, req->offs)
853 : read (req->int1, req->ptr2, req->size); break;
854 case EIO_WRITE: req->result = req->offs >= 0
855 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
856 : write (req->int1, req->ptr2, req->size); break;
857
858 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876 case EIO_CLOSE: req->result = close (req->int1); break;
877 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886 case EIO_READLINK: ALLOC (NAME_MAX);
887 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889 case EIO_SYNC: req->result = 0; sync (); break;
890 case EIO_FSYNC: req->result = fsync (req->int1); break;
891 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893 case EIO_READDIR: eio__scandir (req, self); break;
894
895 case EIO_BUSY:
896#ifdef _WIN32
897 Sleep (req->nv1 * 1000.);
898#else
899 {
900 struct timeval tv;
901
902 tv.tv_sec = req->nv1;
903 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905 req->result = select (0, 0, 0, 0, &tv);
906 }
907#endif
908 break;
909
910 case EIO_UTIME:
911 case EIO_FUTIME:
912 {
913 struct timeval tv[2];
914 struct timeval *times;
915
916 if (req->nv1 != -1. || req->nv2 != -1.)
917 {
918 tv[0].tv_sec = req->nv1;
919 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920 tv[1].tv_sec = req->nv2;
921 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923 times = tv;
924 }
925 else
926 times = 0;
927
928
929 req->result = req->type == EIO_FUTIME
930 ? futimes (req->int1, times)
931 : utimes (req->ptr1, times);
932 }
933
934 case EIO_GROUP:
935 case EIO_NOP:
936 req->result = 0;
937 break;
938
939 case EIO_QUIT:
940 goto quit; 2176 goto quit;
941 2177
942 default: 2178 ETP_EXECUTE (self, req);
943 req->result = -1;
944 break;
945 }
946
947 req->errorno = errno;
948 2179
949 X_LOCK (reslock); 2180 X_LOCK (reslock);
950 2181
951 ++npending; 2182 ++npending;
952 2183
953 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
954 want_poll_cb (); 2185 want_poll_cb ();
955 2186
956 self->req = 0;
957 worker_clear (self); 2187 etp_worker_clear (self);
958 2188
959 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
960 } 2190 }
961 2191
962quit: 2192quit:
2193 free (req);
2194
963 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
964 worker_free (self); 2196 etp_worker_free (self);
965 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
966 2198
967 return 0; 2199 return 0;
968} 2200}
969 2201
970/*****************************************************************************/ 2202/*****************************************************************************/
971 2203
972static void eio_atfork_prepare (void) 2204int ecb_cold
2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
973{ 2206{
974 X_LOCK (wrklock);
975 X_LOCK (reqlock);
976 X_LOCK (reslock);
977#if !HAVE_PREADWRITE 2207#if !HAVE_PREADWRITE
978 X_LOCK (preadwritelock);
979#endif
980#if !HAVE_READDIR_R
981 X_LOCK (readdirlock);
982#endif
983}
984
985static void eio_atfork_parent (void)
986{
987#if !HAVE_READDIR_R
988 X_UNLOCK (readdirlock);
989#endif
990#if !HAVE_PREADWRITE
991 X_UNLOCK (preadwritelock);
992#endif
993 X_UNLOCK (reslock);
994 X_UNLOCK (reqlock);
995 X_UNLOCK (wrklock);
996}
997
998static void eio_atfork_child (void)
999{
1000 eio_req *prv;
1001
1002 while (prv = reqq_shift (&req_queue))
1003 eio_destroy (prv);
1004
1005 while (prv = reqq_shift (&res_queue))
1006 eio_destroy (prv);
1007
1008 while (wrk_first.next != &wrk_first)
1009 {
1010 worker *wrk = wrk_first.next;
1011
1012 if (wrk->req)
1013 eio_destroy (wrk->req);
1014
1015 worker_clear (wrk);
1016 worker_free (wrk);
1017 }
1018
1019 started = 0;
1020 idle = 0;
1021 nreqs = 0;
1022 nready = 0;
1023 npending = 0;
1024
1025 eio_atfork_parent ();
1026}
1027
1028int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029{
1030 want_poll_cb = want_poll;
1031 done_poll_cb = done_poll;
1032
1033#ifdef _WIN32
1034 X_MUTEX_CHECK (wrklock);
1035 X_MUTEX_CHECK (reslock);
1036 X_MUTEX_CHECK (reqlock);
1037 X_MUTEX_CHECK (reqwait);
1038 X_MUTEX_CHECK (preadwritelock); 2208 X_MUTEX_CREATE (preadwritelock);
1039 X_MUTEX_CHECK (readdirlock);
1040
1041 X_COND_CHECK (reqwait);
1042#endif 2209#endif
1043 2210
1044 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2211 return etp_init (want_poll, done_poll);
1045} 2212}
1046 2213
2214ecb_inline void
1047static void eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1048{ 2216{
1049 free (req); 2217 free (req);
1050} 2218}
1051 2219
1052#define REQ(rtype) \ 2220#define REQ(rtype) \
… …
1071 { \ 2239 { \
1072 eio_api_destroy (req); \ 2240 eio_api_destroy (req); \
1073 return 0; \ 2241 return 0; \
1074 } 2242 }
1075 2243
2244static void
2245eio_execute (etp_worker *self, eio_req *req)
2246{
2247#if HAVE_AT
2248 int dirfd;
2249#else
2250 const char *path;
2251#endif
2252
2253 if (ecb_expect_false (EIO_CANCELLED (req)))
2254 {
2255 req->result = -1;
2256 req->errorno = ECANCELED;
2257 return;
2258 }
2259
2260 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2261 {
2262 req->result = -1;
2263 req->errorno = ENOENT;
2264 return;
2265 }
2266
2267 if (req->type >= EIO_OPEN)
2268 {
2269 #if HAVE_AT
2270 dirfd = WD2FD (req->wd);
2271 #else
2272 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2273 #endif
2274 }
2275
2276 switch (req->type)
2277 {
2278 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2279 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2280 break;
2281 case EIO_WD_CLOSE: req->result = 0;
2282 eio_wd_close_sync (req->wd); break;
2283
2284 case EIO_READ: ALLOC (req->size);
2285 req->result = req->offs >= 0
2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
2287 : read (req->int1, req->ptr2, req->size); break;
2288 case EIO_WRITE: req->result = req->offs >= 0
2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2290 : write (req->int1, req->ptr2, req->size); break;
2291
2292 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2293 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2294
2295#if HAVE_AT
2296
2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2298 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2300 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2305
2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2313 case EIO_READLINK: ALLOC (PATH_MAX);
2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2317 case EIO_UTIME:
2318 case EIO_FUTIME:
2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
2341#else
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlink (path ); break;
2353 case EIO_RMDIR: req->result = rmdir (path ); break;
2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363
2364 case EIO_UTIME:
2365 case EIO_FUTIME:
2366 {
2367 struct timeval tv[2];
2368 struct timeval *times;
2369
2370 if (req->nv1 != -1. || req->nv2 != -1.)
2371 {
2372 tv[0].tv_sec = req->nv1;
2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2374 tv[1].tv_sec = req->nv2;
2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2376
2377 times = tv;
2378 }
2379 else
2380 times = 0;
2381
2382 req->result = req->type == EIO_FUTIME
2383 ? futimes (req->int1, times)
2384 : utimes (req->ptr1, times);
2385 }
2386 break;
2387
2388#endif
2389
2390 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2391 {
2392 ALLOC (req->result);
2393 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2394 }
2395 break;
2396
2397 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2398 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2399
2400 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2401 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2402
2403 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2405 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2406
2407 case EIO_CLOSE: req->result = close (req->int1); break;
2408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2409 case EIO_SYNC: req->result = 0; sync (); break;
2410 case EIO_FSYNC: req->result = fsync (req->int1); break;
2411 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2412 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2413 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2414 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2415 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2416 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2417 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2418 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2419
2420 case EIO_READDIR: eio__scandir (req); break;
2421
2422 case EIO_BUSY:
2423#ifdef _WIN32
2424 Sleep (req->nv1 * 1e3);
2425#else
2426 {
2427 struct timeval tv;
2428
2429 tv.tv_sec = req->nv1;
2430 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2431
2432 req->result = select (0, 0, 0, 0, &tv);
2433 }
2434#endif
2435 break;
2436
2437 case EIO_GROUP:
2438 abort (); /* handled in eio_request */
2439
2440 case EIO_NOP:
2441 req->result = 0;
2442 break;
2443
2444 case EIO_CUSTOM:
2445 req->feed (req);
2446 break;
2447
2448 default:
2449 errno = ENOSYS;
2450 req->result = -1;
2451 break;
2452 }
2453
2454 req->errorno = errno;
2455}
2456
1076#ifndef EIO_NO_WRAPPERS 2457#ifndef EIO_NO_WRAPPERS
1077 2458
1078eio_req *eio_nop (int pri, eio_cb cb, void *data) 2459eio_req *eio_nop (int pri, eio_cb cb, void *data)
1079{ 2460{
1080 REQ (EIO_NOP); SEND; 2461 REQ (EIO_NOP); SEND;
… …
1093eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2474eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1094{ 2475{
1095 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2476 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1096} 2477}
1097 2478
2479eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2480{
2481 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2482}
2483
1098eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2484eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1099{ 2485{
1100 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2486 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1101} 2487}
1102 2488
2489eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2492}
2493
2494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2497}
2498
2499eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2500{
2501 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2502}
2503
2504eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2507}
2508
2509eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2512}
2513
2514eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2515{
2516 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2517}
2518
1103eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2519eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1104{ 2520{
1105 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1106} 2522}
1107 2523
… …
1123eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2539eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1124{ 2540{
1125 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2541 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1126} 2542}
1127 2543
2544eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2547}
2548
1128eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2549eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1129{ 2550{
1130 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2551 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1131} 2552}
1132 2553
… …
1138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2559eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1139{ 2560{
1140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1141} 2562}
1142 2563
1143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2564eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1144{ 2565{
1145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2566 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1146} 2567}
1147 2568
1148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2569eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1169{ 2590{
1170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1171} 2592}
1172 2593
1173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2594eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1174{ 2595{
1175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2596 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1176} 2597}
1177 2598
1178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2599eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2615eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1195{ 2616{
1196 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
1197} 2618}
1198 2619
2620eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2621{
2622 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2623}
2624
1199eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2625eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1200{ 2626{
1201 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
1202} 2628}
1203 2629
1204eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2630eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1205{ 2631{
1206 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2632 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1207} 2633}
1208 2634
2635eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2636{
2637 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2638}
2639
1209eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2640eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1210{ 2641{
1211 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2642 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1212} 2643}
1213 2644
1214eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2645eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1215{ 2646{
1216 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2647 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1217} 2648}
1218 2649
1219eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2650eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1220{ 2651{
1221 return eio__1path (EIO_READDIR, path, pri, cb, data); 2652 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1222} 2653}
1223 2654
1224eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2655eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1225{ 2656{
1226 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2657 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1227} 2658}
1228 2659
1229static eio_req * 2660static eio_req *
1230eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2661eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1231{ 2662{
… …
1255eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2686eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1256{ 2687{
1257 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2688 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1258} 2689}
1259 2690
2691eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2692{
2693 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2694}
2695
1260#endif 2696#endif
1261 2697
1262eio_req *eio_grp (eio_cb cb, void *data) 2698eio_req *eio_grp (eio_cb cb, void *data)
1263{ 2699{
1264 const int pri = EIO_PRI_MAX; 2700 const int pri = EIO_PRI_MAX;
… …
1271#undef SEND 2707#undef SEND
1272 2708
1273/*****************************************************************************/ 2709/*****************************************************************************/
1274/* grp functions */ 2710/* grp functions */
1275 2711
2712void
1276void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2713eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1277{ 2714{
1278 grp->int2 = limit; 2715 grp->int2 = limit;
1279 grp->feed = feed; 2716 grp->feed = feed;
1280 2717
1281 grp_try_feed (grp); 2718 grp_try_feed (grp);
1282} 2719}
1283 2720
2721void
1284void eio_grp_limit (eio_req *grp, int limit) 2722eio_grp_limit (eio_req *grp, int limit)
1285{ 2723{
1286 grp->int2 = limit; 2724 grp->int2 = limit;
1287 2725
1288 grp_try_feed (grp); 2726 grp_try_feed (grp);
1289} 2727}
1290 2728
2729void
1291void eio_grp_add (eio_req *grp, eio_req *req) 2730eio_grp_add (eio_req *grp, eio_req *req)
1292{ 2731{
1293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2732 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2733
2734 grp->flags |= EIO_FLAG_GROUPADD;
1294 2735
1295 ++grp->size; 2736 ++grp->size;
1296 req->grp = grp; 2737 req->grp = grp;
1297 2738
1298 req->grp_prev = 0; 2739 req->grp_prev = 0;
… …
1305} 2746}
1306 2747
1307/*****************************************************************************/ 2748/*****************************************************************************/
1308/* misc garbage */ 2749/* misc garbage */
1309 2750
2751eio_ssize_t
1310ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2752eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1311{ 2753{
1312 worker wrk;
1313
1314 wrk.dbuf = 0;
1315
1316 eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
1317
1318 if (wrk.dbuf)
1319 free (wrk.dbuf);
1320} 2755}
1321 2756

Diff Legend

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