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

Comparing libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 2012 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012 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/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
15#ifndef EIO_FINISH 82#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 84#endif
18 85
19#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
22 89
23#ifndef EIO_FEED 90#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 92#endif
26 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
27#ifdef _WIN32 105#ifdef _WIN32
28 106
29 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 196
31#else 197#else
32 198
33# include "config.h"
34# include <sys/time.h> 199 #include <sys/time.h>
35# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
36# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
37# include <utime.h> 237# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 238#endif
44 239
240#if HAVE_SYS_SYSCALL_H
241# include <sys/syscall.h>
242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
45#endif 246#endif
46 247
47#if HAVE_SENDFILE 248#if HAVE_SENDFILE
48# if __linux 249# if __linux
49# include <sys/sendfile.h> 250# include <sys/sendfile.h>
50# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 252# include <sys/socket.h>
52# include <sys/uio.h> 253# include <sys/uio.h>
53# elif __hpux 254# elif __hpux
54# include <sys/socket.h> 255# include <sys/socket.h>
55# elif __solaris /* not yet */ 256# elif __solaris
56# include <sys/sendfile.h> 257# include <sys/sendfile.h>
57# else 258# else
58# error sendfile support requested but not available 259# error sendfile support requested but not available
59# endif 260# endif
60#endif 261#endif
61 262
62/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
64 272
65/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 274#ifndef NAME_MAX
67# define NAME_MAX 4096 275# define NAME_MAX 4096
68#endif 276#endif
69 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
70/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
72 285
73#define dBUF \ 286#define dBUF \
74 char *eio_buf; \
75 X_LOCK (wrklock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \
78 errno = ENOMEM; \ 288 errno = ENOMEM; \
79 if (!eio_buf) \ 289 if (!eio_buf) \
80 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
81 294
82#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
340
341#define ETP_PRI_MIN EIO_PRI_MIN
342#define ETP_PRI_MAX EIO_PRI_MAX
343
344struct etp_worker;
345
346#define ETP_REQ eio_req
347#define ETP_DESTROY(req) eio_destroy (req)
348static int eio_finish (eio_req *req);
349#define ETP_FINISH(req) eio_finish (req)
350static void eio_execute (struct etp_worker *self, eio_req *req);
351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
352
353/*****************************************************************************/
354
355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
356
357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
359tvdiff (struct timeval *tv1, struct timeval *tv2)
360{
361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
363}
364
365static unsigned int started, idle, wanted = 4;
83 366
84static void (*want_poll_cb) (void); 367static void (*want_poll_cb) (void);
85static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
86 369
87static unsigned int max_poll_time = 0; 370static unsigned int max_poll_time; /* reslock */
88static unsigned int max_poll_reqs = 0; 371static unsigned int max_poll_reqs; /* reslock */
89 372
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 373static unsigned int nreqs; /* reqlock */
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 374static unsigned int nready; /* reqlock */
92{ 375static unsigned int npending; /* reqlock */
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95}
96 378
97static unsigned int started, idle, wanted = 4; 379static xmutex_t wrklock;
380static xmutex_t reslock;
381static xmutex_t reqlock;
382static xcond_t reqwait;
383
384#if !HAVE_PREADWRITE
385/*
386 * make our pread/pwrite emulation safe against themselves, but not against
387 * normal read/write by using a mutex. slows down execution a lot,
388 * but that's your problem, not mine.
389 */
390static xmutex_t preadwritelock;
391#endif
392
393typedef struct etp_worker
394{
395 struct tmpbuf tmpbuf;
396
397 /* locked by wrklock */
398 struct etp_worker *prev, *next;
399
400 xthread_t tid;
401
402#ifdef ETP_WORKER_COMMON
403 ETP_WORKER_COMMON
404#endif
405} etp_worker;
406
407static etp_worker wrk_first; /* NOT etp */
408
409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98 411
99/* worker threads management */ 412/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101 413
102typedef struct worker 414static void
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) 415etp_worker_clear (etp_worker *wrk)
118{ 416{
119 if (wrk->dirp) 417}
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124 418
125 if (wrk->dbuf) 419static void ecb_cold
126 { 420etp_worker_free (etp_worker *wrk)
421{
127 free (wrk->dbuf); 422 free (wrk->tmpbuf.ptr);
128 wrk->dbuf = 0;
129 }
130}
131 423
132static void worker_free (worker *wrk)
133{
134 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
136 426
137 free (wrk); 427 free (wrk);
138} 428}
139 429
140static volatile unsigned int nreqs, nready, npending; 430static unsigned int
141static volatile unsigned int max_idle = 4; 431etp_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{ 432{
433 int retval;
434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
435 retval = nreqs;
436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149 return nreqs; 437 return retval;
150} 438}
151 439
152unsigned int eio_nready (void) 440static unsigned int
441etp_nready (void)
153{ 442{
154 unsigned int retval; 443 unsigned int retval;
155 444
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready; 446 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159 448
160 return retval; 449 return retval;
161} 450}
162 451
163unsigned int eio_npending (void) 452static unsigned int
453etp_npending (void)
164{ 454{
165 unsigned int retval; 455 unsigned int retval;
166 456
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending; 458 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170 460
171 return retval; 461 return retval;
172} 462}
173 463
174unsigned int eio_nthreads (void) 464static unsigned int
465etp_nthreads (void)
175{ 466{
176 unsigned int retval; 467 unsigned int retval;
177 468
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started; 470 retval = started;
… …
186 * a somewhat faster data structure might be nice, but 477 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns 478 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation. 479 * per shift, the most expensive operation.
189 */ 480 */
190typedef struct { 481typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 482 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192 int size; 483 int size;
193} reqq; 484} etp_reqq;
194 485
195static reqq req_queue; 486static etp_reqq req_queue;
196static reqq res_queue; 487static etp_reqq res_queue;
197 488
198static int reqq_push (reqq *q, eio_req *req) 489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
501reqq_push (etp_reqq *q, ETP_REQ *req)
199{ 502{
200 int pri = req->pri; 503 int pri = req->pri;
201 req->next = 0; 504 req->next = 0;
202 505
203 if (q->qe[pri]) 506 if (q->qe[pri])
… …
209 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
210 513
211 return q->size++; 514 return q->size++;
212} 515}
213 516
214static eio_req *reqq_shift (reqq *q) 517static ETP_REQ * ecb_noinline
518reqq_shift (etp_reqq *q)
215{ 519{
216 int pri; 520 int pri;
217 521
218 if (!q->size) 522 if (!q->size)
219 return 0; 523 return 0;
220 524
221 --q->size; 525 --q->size;
222 526
223 for (pri = EIO_NUM_PRI; pri--; ) 527 for (pri = ETP_NUM_PRI; pri--; )
224 { 528 {
225 eio_req *req = q->qs[pri]; 529 eio_req *req = q->qs[pri];
226 530
227 if (req) 531 if (req)
228 { 532 {
… …
234 } 538 }
235 539
236 abort (); 540 abort ();
237} 541}
238 542
239static void grp_try_feed (eio_req *grp) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
240{ 545{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 546 X_MUTEX_CREATE (wrklock);
242 { 547 X_MUTEX_CREATE (reslock);
243 int old_len = grp->size; 548 X_MUTEX_CREATE (reqlock);
549 X_COND_CREATE (reqwait);
244 550
245 EIO_FEED (grp); 551 reqq_init (&req_queue);
552 reqq_init (&res_queue);
246 553
247 /* stop if no progress has been made */ 554 wrk_first.next =
248 if (old_len == grp->size) 555 wrk_first.prev = &wrk_first;
249 {
250 grp->feed = 0;
251 break;
252 }
253 }
254}
255 556
256static int eio_finish (eio_req *req); 557 started = 0;
558 idle = 0;
559 nreqs = 0;
560 nready = 0;
561 npending = 0;
257 562
258static int grp_dec (eio_req *grp) 563 want_poll_cb = want_poll;
259{ 564 done_poll_cb = done_poll;
260 --grp->size;
261 565
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; 566 return 0;
270} 567}
271 568
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); 569X_THREAD_PROC (etp_proc);
323 570
324static void start_thread (void) 571static void ecb_cold
572etp_start_thread (void)
325{ 573{
326 worker *wrk = calloc (1, sizeof (worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 575
328 /*TODO*/ 576 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
330 578
331 X_LOCK (wrklock); 579 X_LOCK (wrklock);
332 580
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 581 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 582 {
335 wrk->prev = &wrk_first; 583 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 584 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 585 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 586 wrk_first.next = wrk;
… …
342 free (wrk); 590 free (wrk);
343 591
344 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
345} 593}
346 594
595static void
347static void maybe_start_thread (void) 596etp_maybe_start_thread (void)
348{ 597{
349 if (eio_nthreads () >= wanted) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 599 return;
351 600
352 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 602 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 603 return;
355 604
356 start_thread (); 605 etp_start_thread ();
357} 606}
358 607
359void eio_submit (eio_req *req) 608static void ecb_cold
609etp_end_thread (void)
360{ 610{
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)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
380 612
381 req->type = EIO_QUIT; 613 req->type = -1;
382 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383 615
384 X_LOCK (reqlock); 616 X_LOCK (reqlock);
385 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
386 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
387 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
… …
389 X_LOCK (wrklock); 621 X_LOCK (wrklock);
390 --started; 622 --started;
391 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
392} 624}
393 625
394void eio_set_max_poll_time (double nseconds) 626static 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) 627etp_poll (void)
431{ 628{
432 int maxreqs = max_poll_reqs; 629 unsigned int maxreqs;
630 unsigned int maxtime;
433 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
434 eio_req *req;
435 632
633 X_LOCK (reslock);
634 maxreqs = max_poll_reqs;
635 maxtime = max_poll_time;
636 X_UNLOCK (reslock);
637
436 if (max_poll_time) 638 if (maxtime)
437 gettimeofday (&tv_start, 0); 639 gettimeofday (&tv_start, 0);
438 640
439 for (;;) 641 for (;;)
440 { 642 {
643 ETP_REQ *req;
644
441 maybe_start_thread (); 645 etp_maybe_start_thread ();
442 646
443 X_LOCK (reslock); 647 X_LOCK (reslock);
444 req = reqq_shift (&res_queue); 648 req = reqq_shift (&res_queue);
445 649
446 if (req) 650 if (req)
… …
454 X_UNLOCK (reslock); 658 X_UNLOCK (reslock);
455 659
456 if (!req) 660 if (!req)
457 return 0; 661 return 0;
458 662
663 X_LOCK (reqlock);
459 --nreqs; 664 --nreqs;
665 X_UNLOCK (reqlock);
460 666
461 if (req->type == EIO_GROUP && req->size) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
462 { 668 {
463 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
464 continue; 670 continue;
465 } 671 }
466 else 672 else
467 { 673 {
468 int res = eio_finish (req); 674 int res = ETP_FINISH (req);
469 if (res) 675 if (ecb_expect_false (res))
470 return res; 676 return res;
471 } 677 }
472 678
473 if (maxreqs && !--maxreqs) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
474 break; 680 break;
475 681
476 if (max_poll_time) 682 if (maxtime)
477 { 683 {
478 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
479 685
480 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 686 if (tvdiff (&tv_start, &tv_now) >= maxtime)
481 break; 687 break;
482 } 688 }
483 } 689 }
484 690
485 errno = EAGAIN; 691 errno = EAGAIN;
486 return -1; 692 return -1;
487} 693}
488 694
695static void
696etp_cancel (ETP_REQ *req)
697{
698 req->cancelled = 1;
699
700 eio_grp_cancel (req);
701}
702
703static void
704etp_submit (ETP_REQ *req)
705{
706 req->pri -= ETP_PRI_MIN;
707
708 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
709 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
710
711 if (ecb_expect_false (req->type == EIO_GROUP))
712 {
713 /* I hope this is worth it :/ */
714 X_LOCK (reqlock);
715 ++nreqs;
716 X_UNLOCK (reqlock);
717
718 X_LOCK (reslock);
719
720 ++npending;
721
722 if (!reqq_push (&res_queue, req) && want_poll_cb)
723 want_poll_cb ();
724
725 X_UNLOCK (reslock);
726 }
727 else
728 {
729 X_LOCK (reqlock);
730 ++nreqs;
731 ++nready;
732 reqq_push (&req_queue, req);
733 X_COND_SIGNAL (reqwait);
734 X_UNLOCK (reqlock);
735
736 etp_maybe_start_thread ();
737 }
738}
739
740static void ecb_cold
741etp_set_max_poll_time (double nseconds)
742{
743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
744 max_poll_time = nseconds * EIO_TICKS;
745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
746}
747
748static void ecb_cold
749etp_set_max_poll_reqs (unsigned int maxreqs)
750{
751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
752 max_poll_reqs = maxreqs;
753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
754}
755
756static void ecb_cold
757etp_set_max_idle (unsigned int nthreads)
758{
759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
760 max_idle = nthreads;
761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
762}
763
764static void ecb_cold
765etp_set_idle_timeout (unsigned int seconds)
766{
767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
768 idle_timeout = seconds;
769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
770}
771
772static void ecb_cold
773etp_set_min_parallel (unsigned int nthreads)
774{
775 if (wanted < nthreads)
776 wanted = nthreads;
777}
778
779static void ecb_cold
780etp_set_max_parallel (unsigned int nthreads)
781{
782 if (wanted > nthreads)
783 wanted = nthreads;
784
785 while (started > wanted)
786 etp_end_thread ();
787}
788
789/*****************************************************************************/
790
791static void
792grp_try_feed (eio_req *grp)
793{
794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
795 {
796 grp->flags &= ~EIO_FLAG_GROUPADD;
797
798 EIO_FEED (grp);
799
800 /* stop if no progress has been made */
801 if (!(grp->flags & EIO_FLAG_GROUPADD))
802 {
803 grp->feed = 0;
804 break;
805 }
806 }
807}
808
809static int
810grp_dec (eio_req *grp)
811{
812 --grp->size;
813
814 /* call feeder, if applicable */
815 grp_try_feed (grp);
816
817 /* finish, if done */
818 if (!grp->size && grp->int1)
819 return eio_finish (grp);
820 else
821 return 0;
822}
823
824static void
825eio_destroy (eio_req *req)
826{
827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
829
830 EIO_DESTROY (req);
831}
832
833static int
834eio_finish (eio_req *req)
835{
836 int res = EIO_FINISH (req);
837
838 if (req->grp)
839 {
840 int res2;
841 eio_req *grp = req->grp;
842
843 /* unlink request */
844 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
845 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
846
847 if (grp->grp_first == req)
848 grp->grp_first = req->grp_next;
849
850 res2 = grp_dec (grp);
851
852 if (!res)
853 res = res2;
854 }
855
856 eio_destroy (req);
857
858 return res;
859}
860
861void
862eio_grp_cancel (eio_req *grp)
863{
864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
865 eio_cancel (grp);
866}
867
868void
869eio_cancel (eio_req *req)
870{
871 etp_cancel (req);
872}
873
874void
875eio_submit (eio_req *req)
876{
877 etp_submit (req);
878}
879
880unsigned int
881eio_nreqs (void)
882{
883 return etp_nreqs ();
884}
885
886unsigned int
887eio_nready (void)
888{
889 return etp_nready ();
890}
891
892unsigned int
893eio_npending (void)
894{
895 return etp_npending ();
896}
897
898unsigned int ecb_cold
899eio_nthreads (void)
900{
901 return etp_nthreads ();
902}
903
904void ecb_cold
905eio_set_max_poll_time (double nseconds)
906{
907 etp_set_max_poll_time (nseconds);
908}
909
910void ecb_cold
911eio_set_max_poll_reqs (unsigned int maxreqs)
912{
913 etp_set_max_poll_reqs (maxreqs);
914}
915
916void ecb_cold
917eio_set_max_idle (unsigned int nthreads)
918{
919 etp_set_max_idle (nthreads);
920}
921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
929eio_set_min_parallel (unsigned int nthreads)
930{
931 etp_set_min_parallel (nthreads);
932}
933
934void ecb_cold
935eio_set_max_parallel (unsigned int nthreads)
936{
937 etp_set_max_parallel (nthreads);
938}
939
940int eio_poll (void)
941{
942 return etp_poll ();
943}
944
489/*****************************************************************************/ 945/*****************************************************************************/
490/* work around various missing functions */ 946/* work around various missing functions */
491 947
492#if !HAVE_PREADWRITE 948#if !HAVE_PREADWRITE
949# undef pread
950# undef pwrite
493# define pread eio__pread 951# define pread eio__pread
494# define pwrite eio__pwrite 952# define pwrite eio__pwrite
495 953
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 954static eio_ssize_t
504eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
505{ 956{
506 ssize_t res; 957 eio_ssize_t res;
507 off_t ooffset; 958 off_t ooffset;
508 959
509 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
510 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
511 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
514 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
515 966
516 return res; 967 return res;
517} 968}
518 969
519static ssize_t 970static eio_ssize_t
520eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521{ 972{
522 ssize_t res; 973 eio_ssize_t res;
523 off_t ooffset; 974 off_t ooffset;
524 975
525 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
526 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
527 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
528 res = write (fd, buf, count); 979 res = write (fd, buf, count);
529 lseek (fd, offset, SEEK_SET); 980 lseek (fd, ooffset, SEEK_SET);
530 X_UNLOCK (preadwritelock); 981 X_UNLOCK (preadwritelock);
531 982
532 return res; 983 return res;
533} 984}
534#endif 985#endif
535 986
536#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
537 988
989# undef utimes
538# define utimes(path,times) eio__utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
539# define futimes(fd,times) eio__futimes (fd, times)
540 991
541static int 992static int
542eio__utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
543{ 994{
544 if (times) 995 if (times)
… …
552 } 1003 }
553 else 1004 else
554 return utime (filename, 0); 1005 return utime (filename, 0);
555} 1006}
556 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
557static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
558{ 1017{
559 errno = ENOSYS; 1018 errno = ENOSYS;
560 return -1; 1019 return -1;
561} 1020}
562 1021
563#endif 1022#endif
564 1023
565#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
1025# undef fdatasync
566# define fdatasync fsync 1026# define fdatasync(fd) fsync (fd)
567#endif 1027#endif
1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = EIO_ENOSYS ();
1038#endif
1039
1040 if (res < 0 && errno == ENOSYS && fd >= 0)
1041 sync ();
1042
1043 return res;
1044}
1045
1046/* sync_file_range always needs emulation */
1047static int
1048eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1049{
1050#if HAVE_SYNC_FILE_RANGE
1051 int res;
1052
1053 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1054 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1055 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1056 {
1057 flags = 0
1058 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1059 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1060 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1061 }
1062
1063 res = sync_file_range (fd, offset, nbytes, flags);
1064
1065 if (!res || errno != ENOSYS)
1066 return res;
1067#endif
1068
1069 /* even though we could play tricks with the flags, it's better to always
1070 * call fdatasync, as that matches the expectation of its users best */
1071 return fdatasync (fd);
1072}
1073
1074static int
1075eio__fallocate (int fd, int mode, off_t offset, size_t len)
1076{
1077#if HAVE_LINUX_FALLOCATE
1078 return fallocate (fd, mode, offset, len);
1079#else
1080 return EIO_ENOSYS ();
1081#endif
1082}
568 1083
569#if !HAVE_READAHEAD 1084#if !HAVE_READAHEAD
1085# undef readahead
570# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1086# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571 1087
572static ssize_t 1088static eio_ssize_t
573eio__readahead (int fd, off_t offset, size_t count, worker *self) 1089eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574{ 1090{
575 size_t todo = count; 1091 size_t todo = count;
576 dBUF; 1092 dBUF;
577 1093
578 while (todo > 0) 1094 while (todo > 0)
… …
582 pread (fd, eio_buf, len, offset); 1098 pread (fd, eio_buf, len, offset);
583 offset += len; 1099 offset += len;
584 todo -= len; 1100 todo -= len;
585 } 1101 }
586 1102
587 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
588 return count; 1107 return 0;
589} 1108}
590 1109
591#endif
592
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 1110#endif
623 1111
624/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
625static ssize_t 1113static eio_ssize_t
626eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627{ 1115{
1116 eio_ssize_t written = 0;
628 ssize_t res; 1117 eio_ssize_t res;
629 1118
630 if (!count) 1119 if (!count)
631 return 0; 1120 return 0;
632 1121
1122 for (;;)
1123 {
1124#ifdef __APPLE__
1125# undef HAVE_SENDFILE /* broken, as everything on os x */
1126#endif
633#if HAVE_SENDFILE 1127#if HAVE_SENDFILE
634# if __linux 1128# if __linux
1129 off_t soffset = offset;
635 res = sendfile (ofd, ifd, &offset, count); 1130 res = sendfile (ofd, ifd, &soffset, count);
636 1131
637# elif __freebsd 1132# elif __FreeBSD__
638 /* 1133 /*
639 * Of course, the freebsd sendfile is a dire hack with no thoughts 1134 * Of course, the freebsd sendfile is a dire hack with no thoughts
640 * wasted on making it similar to other I/O functions. 1135 * wasted on making it similar to other I/O functions.
641 */ 1136 */
642 {
643 off_t sbytes; 1137 off_t sbytes;
644 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1138 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645 1139
646 if (res < 0 && sbytes) 1140 #if 0 /* according to the manpage, this is correct, but broken behaviour */
647 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1141 /* freebsd' sendfile will return 0 on success */
1142 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1143 /* not on e.g. EIO or EPIPE - sounds broken */
1144 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648 res = sbytes; 1145 res = sbytes;
649 } 1146 #endif
1147
1148 /* according to source inspection, this is correct, and useful behaviour */
1149 if (sbytes)
1150 res = sbytes;
1151
1152# elif defined __APPLE__
1153 off_t sbytes = count;
1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1155
1156 /* according to the manpage, sbytes is always valid */
1157 if (sbytes)
1158 res = sbytes;
650 1159
651# elif __hpux 1160# elif __hpux
652 res = sendfile (ofd, ifd, offset, count, 0, 0); 1161 res = sendfile (ofd, ifd, offset, count, 0, 0);
653 1162
654# elif __solaris 1163# elif __solaris
655 {
656 struct sendfilevec vec; 1164 struct sendfilevec vec;
657 size_t sbytes; 1165 size_t sbytes;
658 1166
659 vec.sfv_fd = ifd; 1167 vec.sfv_fd = ifd;
660 vec.sfv_flag = 0; 1168 vec.sfv_flag = 0;
661 vec.sfv_off = offset; 1169 vec.sfv_off = offset;
662 vec.sfv_len = count; 1170 vec.sfv_len = count;
663 1171
664 res = sendfilev (ofd, &vec, 1, &sbytes); 1172 res = sendfilev (ofd, &vec, 1, &sbytes);
665 1173
666 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
667 res = sbytes; 1175 res = sbytes;
668 }
669 1176
670# endif 1177# endif
1178
1179#elif defined (_WIN32) && 0
1180 /* does not work, just for documentation of what would need to be done */
1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1182 /* libeio guarantees that the file offset does not change, and windows */
1183 /* has no way to get an independent handle to the same file description */
1184 HANDLE h = TO_SOCKET (ifd);
1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1187
671#else 1188#else
672 res = -1; 1189 res = EIO_ENOSYS ();
673 errno = ENOSYS;
674#endif 1190#endif
675 1191
1192 /* we assume sendfile can copy at least 128mb in one go */
1193 if (res <= 128 * 1024 * 1024)
1194 {
1195 if (res > 0)
1196 written += res;
1197
1198 if (written)
1199 return written;
1200
1201 break;
1202 }
1203 else
1204 {
1205 /* if we requested more, then probably the kernel was lazy */
1206 written += res;
1207 offset += res;
1208 count -= res;
1209
1210 if (!count)
1211 return written;
1212 }
1213 }
1214
676 if (res < 0 1215 if (res < 0
677 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1216 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1217 /* BSDs */
1218#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1219 || errno == ENOTSUP
1220#endif
1221#ifdef EOPNOTSUPP /* windows */
1222 || errno == EOPNOTSUPP /* BSDs */
1223#endif
678#if __solaris 1224#if __solaris
679 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1225 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
680#endif 1226#endif
681 ) 1227 )
682 ) 1228 )
… …
686 1232
687 res = 0; 1233 res = 0;
688 1234
689 while (count) 1235 while (count)
690 { 1236 {
691 ssize_t cnt; 1237 eio_ssize_t cnt;
692 1238
693 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1239 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694 1240
695 if (cnt <= 0) 1241 if (cnt <= 0)
696 { 1242 {
… …
708 1254
709 offset += cnt; 1255 offset += cnt;
710 res += cnt; 1256 res += cnt;
711 count -= cnt; 1257 count -= cnt;
712 } 1258 }
1259
1260 FUBd;
713 } 1261 }
714 1262
715 return res; 1263 return res;
1264}
1265
1266#ifdef PAGESIZE
1267# define eio_pagesize() PAGESIZE
1268#else
1269static intptr_t
1270eio_pagesize (void)
1271{
1272 static intptr_t page;
1273
1274 if (!page)
1275 page = sysconf (_SC_PAGESIZE);
1276
1277 return page;
1278}
1279#endif
1280
1281static void
1282eio_page_align (void **addr, size_t *length)
1283{
1284 intptr_t mask = eio_pagesize () - 1;
1285
1286 /* round down addr */
1287 intptr_t adj = mask & (intptr_t)*addr;
1288
1289 *addr = (void *)((intptr_t)*addr - adj);
1290 *length += adj;
1291
1292 /* round up length */
1293 *length = (*length + mask) & ~mask;
1294}
1295
1296#if !_POSIX_MEMLOCK
1297# define eio__mlockall(a) EIO_ENOSYS ()
1298#else
1299
1300static int
1301eio__mlockall (int flags)
1302{
1303 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1304 extern int mallopt (int, int);
1305 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1306 #endif
1307
1308 if (EIO_MCL_CURRENT != MCL_CURRENT
1309 || EIO_MCL_FUTURE != MCL_FUTURE)
1310 {
1311 flags = 0
1312 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1313 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1314 }
1315
1316 return mlockall (flags);
1317}
1318#endif
1319
1320#if !_POSIX_MEMLOCK_RANGE
1321# define eio__mlock(a,b) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__mlock (void *addr, size_t length)
1326{
1327 eio_page_align (&addr, &length);
1328
1329 return mlock (addr, length);
1330}
1331
1332#endif
1333
1334#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1335# define eio__msync(a,b,c) EIO_ENOSYS ()
1336#else
1337
1338static int
1339eio__msync (void *mem, size_t len, int flags)
1340{
1341 eio_page_align (&mem, &len);
1342
1343 if (EIO_MS_ASYNC != MS_SYNC
1344 || EIO_MS_INVALIDATE != MS_INVALIDATE
1345 || EIO_MS_SYNC != MS_SYNC)
1346 {
1347 flags = 0
1348 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1349 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1350 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1351 }
1352
1353 return msync (mem, len, flags);
1354}
1355
1356#endif
1357
1358static int
1359eio__mtouch (eio_req *req)
1360{
1361 void *mem = req->ptr2;
1362 size_t len = req->size;
1363 int flags = req->int1;
1364
1365 eio_page_align (&mem, &len);
1366
1367 {
1368 intptr_t addr = (intptr_t)mem;
1369 intptr_t end = addr + len;
1370 intptr_t page = eio_pagesize ();
1371
1372 if (addr < end)
1373 if (flags & EIO_MT_MODIFY) /* modify */
1374 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1375 else
1376 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 }
1378
1379 return 0;
1380}
1381
1382/*****************************************************************************/
1383/* requests implemented outside eio_execute, because they are so large */
1384
1385static void
1386eio__lseek (eio_req *req)
1387{
1388 /* this usually gets optimised away completely, or your compiler sucks, */
1389 /* or the whence constants really are not 0, 1, 2 */
1390 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1391 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1392 : req->int2 == EIO_SEEK_END ? SEEK_END
1393 : req->int2;
1394
1395 req->offs = lseek (req->int1, req->offs, whence);
1396 req->result = req->offs == (off_t)-1 ? -1 : 0;
1397}
1398
1399/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1400static int
1401eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1402{
1403 const char *rel = path;
1404 char *res;
1405 char *tmp1, *tmp2;
1406#if SYMLOOP_MAX > 32
1407 int symlinks = SYMLOOP_MAX;
1408#else
1409 int symlinks = 32;
1410#endif
1411
1412 errno = EINVAL;
1413 if (!rel)
1414 return -1;
1415
1416 errno = ENOENT;
1417 if (!*rel)
1418 return -1;
1419
1420 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1421 tmp1 = res + PATH_MAX;
1422 tmp2 = tmp1 + PATH_MAX;
1423
1424#if 0 /* disabled, the musl way to do things is just too racy */
1425#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1426 /* on linux we may be able to ask the kernel */
1427 {
1428 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1429
1430 if (fd >= 0)
1431 {
1432 sprintf (tmp1, "/proc/self/fd/%d", fd);
1433 req->result = readlink (tmp1, res, PATH_MAX);
1434 close (fd);
1435
1436 /* here we should probably stat the open file and the disk file, to make sure they still match */
1437
1438 if (req->result > 0)
1439 goto done;
1440 }
1441 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1442 return;
1443 }
1444#endif
1445#endif
1446
1447 if (*rel != '/')
1448 {
1449 int len;
1450
1451 errno = ENOENT;
1452 if (wd == EIO_INVALID_WD)
1453 return -1;
1454
1455 if (wd == EIO_CWD)
1456 {
1457 if (!getcwd (res, PATH_MAX))
1458 return -1;
1459
1460 len = strlen (res);
1461 }
1462 else
1463 memcpy (res, wd->str, len = wd->len);
1464
1465 if (res [1]) /* only use if not / */
1466 res += len;
1467 }
1468
1469 while (*rel)
1470 {
1471 eio_ssize_t len, linklen;
1472 const char *beg = rel;
1473
1474 while (*rel && *rel != '/')
1475 ++rel;
1476
1477 len = rel - beg;
1478
1479 if (!len) /* skip slashes */
1480 {
1481 ++rel;
1482 continue;
1483 }
1484
1485 if (beg [0] == '.')
1486 {
1487 if (len == 1)
1488 continue; /* . - nop */
1489
1490 if (beg [1] == '.' && len == 2)
1491 {
1492 /* .. - back up one component, if possible */
1493
1494 while (res != tmpbuf->ptr)
1495 if (*--res == '/')
1496 break;
1497
1498 continue;
1499 }
1500 }
1501
1502 errno = ENAMETOOLONG;
1503 if (res + 1 + len + 1 >= tmp1)
1504 return -1;
1505
1506 /* copy one component */
1507 *res = '/';
1508 memcpy (res + 1, beg, len);
1509
1510 /* zero-terminate, for readlink */
1511 res [len + 1] = 0;
1512
1513 /* now check if it's a symlink */
1514 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1515
1516 if (linklen < 0)
1517 {
1518 if (errno != EINVAL)
1519 return -1;
1520
1521 /* it's a normal directory. hopefully */
1522 res += len + 1;
1523 }
1524 else
1525 {
1526 /* yay, it was a symlink - build new path in tmp2 */
1527 int rellen = strlen (rel);
1528
1529 errno = ENAMETOOLONG;
1530 if (linklen + 1 + rellen >= PATH_MAX)
1531 return -1;
1532
1533 errno = ELOOP;
1534 if (!--symlinks)
1535 return -1;
1536
1537 if (*tmp1 == '/')
1538 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1539
1540 /* we need to be careful, as rel might point into tmp2 already */
1541 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1542 tmp2 [linklen] = '/';
1543 memcpy (tmp2, tmp1, linklen);
1544
1545 rel = tmp2;
1546 }
1547 }
1548
1549 /* special case for the lone root path */
1550 if (res == tmpbuf->ptr)
1551 *res++ = '/';
1552
1553 return res - (char *)tmpbuf->ptr;
1554}
1555
1556static signed char
1557eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1558{
1559 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1560 : a->inode < b->inode ? -1
1561 : a->inode > b->inode ? 1
1562 : 0;
1563}
1564
1565#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1566
1567#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1568#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1569
1570static void
1571eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1572{
1573 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1574 unsigned char *bit = bits;
1575
1576 assert (CHAR_BIT == 8);
1577 assert (sizeof (eio_dirent) * 8 < 256);
1578 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1579 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1580
1581 if (size <= EIO_SORT_FAST)
1582 return;
1583
1584 /* first prepare an array of bits to test in our radix sort */
1585 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1586 /* inode_bits must contain all inodes ORed together */
1587 /* which is used to skip bits that are 0 everywhere, which is very common */
1588 {
1589 eio_ino_t endianness;
1590 int i, j;
1591
1592 /* we store the byte offset of byte n into byte n of "endianness" */
1593 for (i = 0; i < sizeof (eio_ino_t); ++i)
1594 ((unsigned char *)&endianness)[i] = i;
1595
1596 *bit++ = 0;
1597
1598 for (i = 0; i < sizeof (eio_ino_t); ++i)
1599 {
1600 /* shifting off the byte offsets out of "endianness" */
1601 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1602 endianness >>= 8;
1603
1604 for (j = 0; j < 8; ++j)
1605 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1606 *bit++ = offs + j;
1607 }
1608
1609 for (j = 0; j < 8; ++j)
1610 if (score_bits & (1 << j))
1611 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1612 }
1613
1614 /* now actually do the sorting (a variant of MSD radix sort) */
1615 {
1616 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1617 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1618 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1619 int stk_idx = 0;
1620
1621 base_stk [stk_idx] = dents;
1622 end_stk [stk_idx] = dents + size;
1623 bit_stk [stk_idx] = bit - 1;
1624
1625 do
1626 {
1627 base = base_stk [stk_idx];
1628 end = end_stk [stk_idx];
1629 bit = bit_stk [stk_idx];
1630
1631 for (;;)
1632 {
1633 unsigned char O = *bit >> 3;
1634 unsigned char M = 1 << (*bit & 7);
1635
1636 eio_dirent *a = base;
1637 eio_dirent *b = end;
1638
1639 if (b - a < EIO_SORT_CUTOFF)
1640 break;
1641
1642 /* now bit-partition the array on the bit */
1643 /* this ugly asymmetric loop seems to perform much better than typical */
1644 /* partition algos found in the literature */
1645 do
1646 if (!(((unsigned char *)a)[O] & M))
1647 ++a;
1648 else if (!(((unsigned char *)--b)[O] & M))
1649 {
1650 eio_dirent tmp = *a; *a = *b; *b = tmp;
1651 ++a;
1652 }
1653 while (b > a);
1654
1655 /* next bit, or stop, if no bits left in this path */
1656 if (!*--bit)
1657 break;
1658
1659 base_stk [stk_idx] = a;
1660 end_stk [stk_idx] = end;
1661 bit_stk [stk_idx] = bit;
1662 ++stk_idx;
1663
1664 end = a;
1665 }
1666 }
1667 while (stk_idx--);
1668 }
1669}
1670
1671static void
1672eio_dent_insertion_sort (eio_dirent *dents, int size)
1673{
1674 /* first move the smallest element to the front, to act as a sentinel */
1675 {
1676 int i;
1677 eio_dirent *min = dents;
1678
1679 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1680 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1681 if (EIO_DENT_CMP (dents [i], <, *min))
1682 min = &dents [i];
1683
1684 /* swap elements 0 and j (minimum) */
1685 {
1686 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1687 }
1688 }
1689
1690 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1691 {
1692 eio_dirent *i, *j;
1693
1694 for (i = dents + 1; i < dents + size; ++i)
1695 {
1696 eio_dirent value = *i;
1697
1698 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1699 j [1] = j [0];
1700
1701 j [1] = value;
1702 }
1703 }
1704}
1705
1706static void
1707eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1708{
1709 if (size <= 1)
1710 return; /* our insertion sort relies on size > 0 */
1711
1712 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1713 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1714 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1715
1716 /* use an insertion sort at the end, or for small arrays, */
1717 /* as insertion sort is more efficient for small partitions */
1718 eio_dent_insertion_sort (dents, size);
716} 1719}
717 1720
718/* read a full directory */ 1721/* read a full directory */
719static void 1722static void
720eio__scandir (eio_req *req, worker *self) 1723eio__scandir (eio_req *req, etp_worker *self)
721{ 1724{
1725 char *name, *names;
1726 int namesalloc = 4096 - sizeof (void *) * 4;
1727 int namesoffs = 0;
1728 int flags = req->int1;
1729 eio_dirent *dents = 0;
1730 int dentalloc = 128;
1731 int dentoffs = 0;
1732 eio_ino_t inode_bits = 0;
1733#ifdef _WIN32
1734 HANDLE dirp;
1735 WIN32_FIND_DATA entp;
1736#else
722 DIR *dirp; 1737 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; 1738 EIO_STRUCT_DIRENT *entp;
729 char *name, *names; 1739#endif
730 int memlen = 4096;
731 int memofs = 0;
732 int res = 0;
733 1740
734 X_LOCK (wrklock); 1741 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 1742
741 if (dirp && u && names) 1743 if (!(flags & EIO_READDIR_DENTS))
742 for (;;) 1744 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1745
1746#ifdef _WIN32
1747 {
1748 int len = strlen ((const char *)req->ptr1);
1749 char *path = malloc (MAX_PATH);
1750 const char *fmt;
1751 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1752
1753 if (!len)
1754 fmt = "./*";
1755 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1756 fmt = "%s*";
1757 else
1758 fmt = "%s/*";
1759
1760 _snprintf (path, MAX_PATH, fmt, reqpath);
1761 dirp = FindFirstFile (path, &entp);
1762 free (path);
1763
1764 if (dirp == INVALID_HANDLE_VALUE)
1765 {
1766 /* should steal _dosmaperr */
1767 switch (GetLastError ())
1768 {
1769 case ERROR_FILE_NOT_FOUND:
1770 req->result = 0;
1771 break;
1772
1773 case ERROR_INVALID_NAME:
1774 case ERROR_PATH_NOT_FOUND:
1775 case ERROR_NO_MORE_FILES:
1776 errno = ENOENT;
1777 break;
1778
1779 case ERROR_NOT_ENOUGH_MEMORY:
1780 errno = ENOMEM;
1781 break;
1782
1783 default:
1784 errno = EINVAL;
1785 break;
1786 }
1787
1788 return;
1789 }
1790 }
1791#else
1792 #if HAVE_AT
1793 if (req->wd)
743 { 1794 {
1795 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1796
1797 if (fd < 0)
1798 return;
1799
1800 dirp = fdopendir (fd);
1801
1802 if (!dirp)
1803 close (fd);
1804 }
1805 else
1806 dirp = opendir (req->ptr1);
1807 #else
1808 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1809 #endif
1810
1811 if (!dirp)
1812 return;
1813#endif
1814
1815 if (req->flags & EIO_FLAG_PTR1_FREE)
1816 free (req->ptr1);
1817
1818 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1819 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1820 req->ptr2 = names = malloc (namesalloc);
1821
1822 if (!names || (flags && !dents))
1823 return;
1824
1825 for (;;)
1826 {
1827 int done;
1828
1829#ifdef _WIN32
1830 done = !dirp;
1831#else
744 errno = 0; 1832 errno = 0;
745 readdir_r (dirp, &u->d, &entp); 1833 entp = readdir (dirp);
1834 done = !entp;
1835#endif
746 1836
1837 if (done)
1838 {
1839#ifndef _WIN32
1840 int old_errno = errno;
1841 closedir (dirp);
1842 errno = old_errno;
1843
747 if (!entp) 1844 if (errno)
1845 break;
1846#endif
1847
1848 /* sort etc. */
1849 req->int1 = flags;
1850 req->result = dentoffs;
1851
1852 if (flags & EIO_READDIR_STAT_ORDER)
1853 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1854 else if (flags & EIO_READDIR_DIRS_FIRST)
1855 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1856 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1857 else
1858 {
1859 /* in this case, all is known, and we just put dirs first and sort them */
1860 eio_dirent *oth = dents + dentoffs;
1861 eio_dirent *dir = dents;
1862
1863 /* now partition dirs to the front, and non-dirs to the back */
1864 /* by walking from both sides and swapping if necessary */
1865 while (oth > dir)
1866 {
1867 if (dir->type == EIO_DT_DIR)
1868 ++dir;
1869 else if ((--oth)->type == EIO_DT_DIR)
1870 {
1871 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1872
1873 ++dir;
1874 }
1875 }
1876
1877 /* now sort the dirs only (dirs all have the same score) */
1878 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1879 }
1880
748 break; 1881 break;
1882 }
749 1883
750 name = entp->d_name; 1884 /* now add the entry to our list(s) */
1885 name = D_NAME (entp);
751 1886
1887 /* skip . and .. entries */
752 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1888 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753 { 1889 {
754 int len = strlen (name) + 1; 1890 int len = D_NAMLEN (entp) + 1;
755 1891
756 res++; 1892 while (ecb_expect_false (namesoffs + len > namesalloc))
757
758 while (memofs + len > memlen)
759 { 1893 {
760 memlen *= 2; 1894 namesalloc *= 2;
761 X_LOCK (wrklock);
762 req->ptr2 = names = realloc (names, memlen); 1895 req->ptr2 = names = realloc (names, namesalloc);
763 X_UNLOCK (wrklock);
764 1896
765 if (!names) 1897 if (!names)
1898 break;
1899 }
1900
1901 memcpy (names + namesoffs, name, len);
1902
1903 if (dents)
1904 {
1905 struct eio_dirent *ent;
1906
1907 if (ecb_expect_false (dentoffs == dentalloc))
1908 {
1909 dentalloc *= 2;
1910 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1911
1912 if (!dents)
766 break; 1913 break;
767 } 1914 }
768 1915
769 memcpy (names + memofs, name, len); 1916 ent = dents + dentoffs;
770 memofs += len; 1917
1918 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1919 ent->namelen = len - 1;
1920 ent->inode = D_INO (entp);
1921
1922 inode_bits |= ent->inode;
1923
1924 switch (D_TYPE (entp))
1925 {
1926 default:
1927 ent->type = EIO_DT_UNKNOWN;
1928 flags |= EIO_READDIR_FOUND_UNKNOWN;
1929 break;
1930
1931 #ifdef DT_FIFO
1932 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1933 #endif
1934 #ifdef DT_CHR
1935 case DT_CHR: ent->type = EIO_DT_CHR; break;
1936 #endif
1937 #ifdef DT_MPC
1938 case DT_MPC: ent->type = EIO_DT_MPC; break;
1939 #endif
1940 #ifdef DT_DIR
1941 case DT_DIR: ent->type = EIO_DT_DIR; break;
1942 #endif
1943 #ifdef DT_NAM
1944 case DT_NAM: ent->type = EIO_DT_NAM; break;
1945 #endif
1946 #ifdef DT_BLK
1947 case DT_BLK: ent->type = EIO_DT_BLK; break;
1948 #endif
1949 #ifdef DT_MPB
1950 case DT_MPB: ent->type = EIO_DT_MPB; break;
1951 #endif
1952 #ifdef DT_REG
1953 case DT_REG: ent->type = EIO_DT_REG; break;
1954 #endif
1955 #ifdef DT_NWK
1956 case DT_NWK: ent->type = EIO_DT_NWK; break;
1957 #endif
1958 #ifdef DT_CMP
1959 case DT_CMP: ent->type = EIO_DT_CMP; break;
1960 #endif
1961 #ifdef DT_LNK
1962 case DT_LNK: ent->type = EIO_DT_LNK; break;
1963 #endif
1964 #ifdef DT_SOCK
1965 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1966 #endif
1967 #ifdef DT_DOOR
1968 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1969 #endif
1970 #ifdef DT_WHT
1971 case DT_WHT: ent->type = EIO_DT_WHT; break;
1972 #endif
1973 }
1974
1975 ent->score = 7;
1976
1977 if (flags & EIO_READDIR_DIRS_FIRST)
1978 {
1979 if (ent->type == EIO_DT_UNKNOWN)
1980 {
1981 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1982 ent->score = 1;
1983 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1984 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1985 }
1986 else if (ent->type == EIO_DT_DIR)
1987 ent->score = 0;
1988 }
771 } 1989 }
1990
1991 namesoffs += len;
1992 ++dentoffs;
772 } 1993 }
773 1994
774 if (errno) 1995 if (EIO_CANCELLED (req))
775 res = -1; 1996 {
1997 errno = ECANCELED;
1998 break;
1999 }
2000
2001#ifdef _WIN32
2002 if (!FindNextFile (dirp, &entp))
2003 {
2004 FindClose (dirp);
2005 dirp = 0;
2006 }
2007#endif
2008 }
2009}
2010
2011/*****************************************************************************/
2012/* working directory stuff */
2013/* various deficiencies in the posix 2008 api force us to */
2014/* keep the absolute path in string form at all times */
2015/* fuck yeah. */
2016
2017#if !HAVE_AT
2018
2019/* a bit like realpath, but usually faster because it doesn'T have to return */
2020/* an absolute or canonical path */
2021static const char *
2022wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2023{
2024 if (!wd || *path == '/')
2025 return path;
2026
2027 if (path [0] == '.' && !path [1])
2028 return wd->str;
2029
776 2030 {
777 req->result = res; 2031 int l1 = wd->len;
2032 int l2 = strlen (path);
2033
2034 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2035
2036 memcpy (res, wd->str, l1);
2037 res [l1] = '/';
2038 memcpy (res + l1 + 1, path, l2 + 1);
2039
2040 return res;
2041 }
778} 2042}
2043
2044#endif
2045
2046static eio_wd
2047eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2048{
2049 int fd;
2050 eio_wd res;
2051 int len = eio__realpath (tmpbuf, wd, path);
2052
2053 if (len < 0)
2054 return EIO_INVALID_WD;
2055
2056#if HAVE_AT
2057 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2058
2059 if (fd < 0)
2060 return EIO_INVALID_WD;
2061#endif
2062
2063 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2064
2065#if HAVE_AT
2066 res->fd = fd;
2067#endif
2068
2069 res->len = len;
2070 memcpy (res->str, tmpbuf->ptr, len);
2071 res->str [len] = 0;
2072
2073 return res;
2074}
2075
2076eio_wd
2077eio_wd_open_sync (eio_wd wd, const char *path)
2078{
2079 struct tmpbuf tmpbuf = { 0 };
2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
2082
2083 return wd;
2084}
2085
2086void
2087eio_wd_close_sync (eio_wd wd)
2088{
2089 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2090 {
2091 #if HAVE_AT
2092 close (wd->fd);
2093 #endif
2094 free (wd);
2095 }
2096}
2097
2098#if HAVE_AT
2099
2100/* they forgot these */
2101
2102static int
2103eio__truncateat (int dirfd, const char *path, off_t length)
2104{
2105 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = ftruncate (fd, length);
2112 close (fd);
2113 return res;
2114}
2115
2116static int
2117eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2118{
2119 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2120 int res;
2121
2122 if (fd < 0)
2123 return fd;
2124
2125 res = fstatvfs (fd, buf);
2126 close (fd);
2127 return res;
2128
2129}
2130
2131#endif
779 2132
780/*****************************************************************************/ 2133/*****************************************************************************/
781 2134
782#define ALLOC(len) \ 2135#define ALLOC(len) \
783 if (!req->ptr2) \ 2136 if (!req->ptr2) \
… …
792 req->result = -1; \ 2145 req->result = -1; \
793 break; \ 2146 break; \
794 } \ 2147 } \
795 } 2148 }
796 2149
2150static void ecb_noinline ecb_cold
2151etp_proc_init (void)
2152{
2153#if HAVE_PRCTL_SET_NAME
2154 /* provide a more sensible "thread name" */
2155 char name[16 + 1];
2156 const int namelen = sizeof (name) - 1;
2157 int len;
2158
2159 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2160 name [namelen] = 0;
2161 len = strlen (name);
2162 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2163 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2164#endif
2165}
2166
797X_THREAD_PROC (eio_proc) 2167X_THREAD_PROC (etp_proc)
798{ 2168{
799 eio_req *req; 2169 ETP_REQ *req;
800 struct timespec ts; 2170 struct timespec ts;
801 worker *self = (worker *)thr_arg; 2171 etp_worker *self = (etp_worker *)thr_arg;
802 2172
2173 etp_proc_init ();
2174
803 /* try to distribute timeouts somewhat randomly */ 2175 /* try to distribute timeouts somewhat evenly */
804 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2176 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805 2177
806 for (;;) 2178 for (;;)
807 { 2179 {
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2180 ts.tv_sec = 0;
809 2181
810 X_LOCK (reqlock); 2182 X_LOCK (reqlock);
811 2183
812 for (;;) 2184 for (;;)
813 { 2185 {
814 self->req = req = reqq_shift (&req_queue); 2186 req = reqq_shift (&req_queue);
815 2187
816 if (req) 2188 if (req)
817 break; 2189 break;
818 2190
2191 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2192 {
2193 X_UNLOCK (reqlock);
2194 X_LOCK (wrklock);
2195 --started;
2196 X_UNLOCK (wrklock);
2197 goto quit;
2198 }
2199
819 ++idle; 2200 ++idle;
820 2201
821 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2202 if (idle <= max_idle)
2203 /* we are allowed to idle, so do so without any timeout */
2204 X_COND_WAIT (reqwait, reqlock);
2205 else
822 { 2206 {
823 if (idle > max_idle) 2207 /* initialise timeout once */
824 { 2208 if (!ts.tv_sec)
825 --idle; 2209 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 2210
833 /* we are allowed to idle, so do so without any timeout */
834 X_COND_WAIT (reqwait, reqlock); 2211 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
835 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2212 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
836 } 2213 }
837 2214
838 --idle; 2215 --idle;
839 } 2216 }
840 2217
841 --nready; 2218 --nready;
842 2219
843 X_UNLOCK (reqlock); 2220 X_UNLOCK (reqlock);
844 2221
845 errno = 0; /* strictly unnecessary */
846
847 if (!EIO_CANCELLED (req))
848 switch (req->type) 2222 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; 2223 goto quit;
941 2224
942 default: 2225 ETP_EXECUTE (self, req);
943 req->result = -1;
944 break;
945 }
946
947 req->errorno = errno;
948 2226
949 X_LOCK (reslock); 2227 X_LOCK (reslock);
950 2228
951 ++npending; 2229 ++npending;
952 2230
953 if (!reqq_push (&res_queue, req) && want_poll_cb) 2231 if (!reqq_push (&res_queue, req) && want_poll_cb)
954 want_poll_cb (); 2232 want_poll_cb ();
955 2233
956 self->req = 0;
957 worker_clear (self); 2234 etp_worker_clear (self);
958 2235
959 X_UNLOCK (reslock); 2236 X_UNLOCK (reslock);
960 } 2237 }
961 2238
962quit: 2239quit:
2240 free (req);
2241
963 X_LOCK (wrklock); 2242 X_LOCK (wrklock);
964 worker_free (self); 2243 etp_worker_free (self);
965 X_UNLOCK (wrklock); 2244 X_UNLOCK (wrklock);
966
967 return 0;
968} 2245}
969 2246
970/*****************************************************************************/ 2247/*****************************************************************************/
971 2248
972static void eio_atfork_prepare (void) 2249int ecb_cold
2250eio_init (void (*want_poll)(void), void (*done_poll)(void))
973{ 2251{
974 X_LOCK (wrklock);
975 X_LOCK (reqlock);
976 X_LOCK (reslock);
977#if !HAVE_PREADWRITE 2252#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); 2253 X_MUTEX_CREATE (preadwritelock);
1039 X_MUTEX_CHECK (readdirlock);
1040
1041 X_COND_CHECK (reqwait);
1042#endif 2254#endif
1043 2255
1044 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2256 return etp_init (want_poll, done_poll);
1045} 2257}
1046 2258
2259ecb_inline void
1047static void eio_api_destroy (eio_req *req) 2260eio_api_destroy (eio_req *req)
1048{ 2261{
1049 free (req); 2262 free (req);
1050} 2263}
1051 2264
1052#define REQ(rtype) \ 2265#define REQ(rtype) \
… …
1071 { \ 2284 { \
1072 eio_api_destroy (req); \ 2285 eio_api_destroy (req); \
1073 return 0; \ 2286 return 0; \
1074 } 2287 }
1075 2288
2289static void
2290eio_execute (etp_worker *self, eio_req *req)
2291{
2292#if HAVE_AT
2293 int dirfd;
2294#else
2295 const char *path;
2296#endif
2297
2298 if (ecb_expect_false (EIO_CANCELLED (req)))
2299 {
2300 req->result = -1;
2301 req->errorno = ECANCELED;
2302 return;
2303 }
2304
2305 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2306 {
2307 req->result = -1;
2308 req->errorno = ENOENT;
2309 return;
2310 }
2311
2312 if (req->type >= EIO_OPEN)
2313 {
2314 #if HAVE_AT
2315 dirfd = WD2FD (req->wd);
2316 #else
2317 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2318 #endif
2319 }
2320
2321 switch (req->type)
2322 {
2323 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2324 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2325 break;
2326 case EIO_WD_CLOSE: req->result = 0;
2327 eio_wd_close_sync (req->wd); break;
2328
2329 case EIO_SEEK: eio__lseek (req); break;
2330 case EIO_READ: ALLOC (req->size);
2331 req->result = req->offs >= 0
2332 ? pread (req->int1, req->ptr2, req->size, req->offs)
2333 : read (req->int1, req->ptr2, req->size); break;
2334 case EIO_WRITE: req->result = req->offs >= 0
2335 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2336 : write (req->int1, req->ptr2, req->size); break;
2337
2338 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2339 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2340
2341#if HAVE_AT
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2347 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2348 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2349 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2350 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2353 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2354 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2445
2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2448
2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2452
2453 case EIO_CLOSE: req->result = close (req->int1); break;
2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2455 case EIO_SYNC: req->result = 0; sync (); break;
2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2465
2466 case EIO_READDIR: eio__scandir (req, self); break;
2467
2468 case EIO_BUSY:
2469#ifdef _WIN32
2470 Sleep (req->nv1 * 1e3);
2471#else
2472 {
2473 struct timeval tv;
2474
2475 tv.tv_sec = req->nv1;
2476 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2477
2478 req->result = select (0, 0, 0, 0, &tv);
2479 }
2480#endif
2481 break;
2482
2483 case EIO_GROUP:
2484 abort (); /* handled in eio_request */
2485
2486 case EIO_NOP:
2487 req->result = 0;
2488 break;
2489
2490 case EIO_CUSTOM:
2491 req->feed (req);
2492 break;
2493
2494 default:
2495 req->result = EIO_ENOSYS ();
2496 break;
2497 }
2498
2499 req->errorno = errno;
2500}
2501
2502#ifndef EIO_NO_WRAPPERS
2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
1076eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077{ 2515{
1078 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
1079} 2517}
1080 2518
… …
1091eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2529eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092{ 2530{
1093 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2531 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1094} 2532}
1095 2533
2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2537}
2538
1096eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097{ 2540{
1098 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1099} 2542}
1100 2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2555{
2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2557}
2558
2559eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2560{
2561 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2562}
2563
2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2565{
2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2567}
2568
2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2572}
2573
1101eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1102{ 2575{
1103 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1104} 2577}
1105 2578
1106eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1107{ 2580{
1108 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1109} 2582}
1110 2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
1111eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1112{ 2590{
1113 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1114} 2592}
1115 2593
… …
1121eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2599eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1122{ 2600{
1123 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2601 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1124} 2602}
1125 2603
2604eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2605{
2606 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2607}
2608
1126eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2609eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1127{ 2610{
1128 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2611 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1129} 2612}
1130 2613
… …
1136eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1137{ 2620{
1138 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1139} 2622}
1140 2623
1141eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1142{ 2625{
1143 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1144} 2627}
1145 2628
1146eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1166eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1167{ 2650{
1168 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1169} 2652}
1170 2653
1171eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1172{ 2655{
1173 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1174} 2657}
1175 2658
1176eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1192eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2675eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1193{ 2676{
1194 return eio__1path (EIO_READLINK, path, pri, cb, data); 2677 return eio__1path (EIO_READLINK, path, pri, cb, data);
1195} 2678}
1196 2679
2680eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2681{
2682 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2683}
2684
1197eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2685eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1198{ 2686{
1199 return eio__1path (EIO_STAT, path, pri, cb, data); 2687 return eio__1path (EIO_STAT, path, pri, cb, data);
1200} 2688}
1201 2689
1202eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2690eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1203{ 2691{
1204 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2692 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1205} 2693}
1206 2694
2695eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2696{
2697 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2698}
2699
1207eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2700eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1208{ 2701{
1209 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2702 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1210} 2703}
1211 2704
1212eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2705eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1213{ 2706{
1214 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2707 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1215} 2708}
1216 2709
1217eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2710eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1218{ 2711{
1219 return eio__1path (EIO_READDIR, path, pri, cb, data); 2712 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1220} 2713}
1221 2714
1222eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2715eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1223{ 2716{
1224 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2717 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1225} 2718}
1226 2719
1227static eio_req * 2720static eio_req *
1228eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2721eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1229{ 2722{
… …
1253eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2746eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1254{ 2747{
1255 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2748 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256} 2749}
1257 2750
2751eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2752{
2753 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2754}
2755
2756#endif
2757
1258eio_req *eio_grp (eio_cb cb, void *data) 2758eio_req *eio_grp (eio_cb cb, void *data)
1259{ 2759{
1260 const int pri = EIO_PRI_MAX; 2760 const int pri = EIO_PRI_MAX;
1261 2761
1262 REQ (EIO_GROUP); SEND; 2762 REQ (EIO_GROUP); SEND;
… …
1267#undef SEND 2767#undef SEND
1268 2768
1269/*****************************************************************************/ 2769/*****************************************************************************/
1270/* grp functions */ 2770/* grp functions */
1271 2771
2772void
1272void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2773eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1273{ 2774{
1274 grp->int2 = limit; 2775 grp->int2 = limit;
1275 grp->feed = feed; 2776 grp->feed = feed;
1276 2777
1277 grp_try_feed (grp); 2778 grp_try_feed (grp);
1278} 2779}
1279 2780
2781void
1280void eio_grp_limit (eio_req *grp, int limit) 2782eio_grp_limit (eio_req *grp, int limit)
1281{ 2783{
1282 grp->int2 = limit; 2784 grp->int2 = limit;
1283 2785
1284 grp_try_feed (grp); 2786 grp_try_feed (grp);
1285} 2787}
1286 2788
2789void
1287void eio_grp_add (eio_req *grp, eio_req *req) 2790eio_grp_add (eio_req *grp, eio_req *req)
1288{ 2791{
1289 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2792 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2793
2794 grp->flags |= EIO_FLAG_GROUPADD;
1290 2795
1291 ++grp->size; 2796 ++grp->size;
1292 req->grp = grp; 2797 req->grp = grp;
1293 2798
1294 req->grp_prev = 0; 2799 req->grp_prev = 0;
… …
1301} 2806}
1302 2807
1303/*****************************************************************************/ 2808/*****************************************************************************/
1304/* misc garbage */ 2809/* misc garbage */
1305 2810
2811eio_ssize_t
1306ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1307{ 2813{
1308 worker wrk;
1309
1310 wrk.dbuf = 0;
1311
1312 eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
1313
1314 if (wrk.dbuf)
1315 free (wrk.dbuf);
1316} 2815}
1317 2816

Diff Legend

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