1 |
root |
1.14 |
/* |
2 |
|
|
* libeio implementation |
3 |
|
|
* |
4 |
root |
1.61 |
* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> |
5 |
root |
1.14 |
* 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 |
root |
1.65 |
#ifndef _WIN32 |
41 |
|
|
# include "config.h" |
42 |
|
|
#endif |
43 |
|
|
|
44 |
root |
1.1 |
#include "eio.h" |
45 |
root |
1.29 |
|
46 |
|
|
#ifdef EIO_STACKSIZE |
47 |
|
|
# define XTHREAD_STACKSIZE EIO_STACKSIZE |
48 |
|
|
#endif |
49 |
root |
1.1 |
#include "xthread.h" |
50 |
|
|
|
51 |
|
|
#include <errno.h> |
52 |
|
|
#include <stddef.h> |
53 |
|
|
#include <stdlib.h> |
54 |
root |
1.3 |
#include <string.h> |
55 |
root |
1.1 |
#include <errno.h> |
56 |
|
|
#include <sys/types.h> |
57 |
|
|
#include <sys/stat.h> |
58 |
root |
1.51 |
#include <sys/statvfs.h> |
59 |
root |
1.1 |
#include <limits.h> |
60 |
|
|
#include <fcntl.h> |
61 |
root |
1.8 |
#include <assert.h> |
62 |
root |
1.1 |
|
63 |
root |
1.66 |
/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
64 |
|
|
/* intptr_t only comes form stdint.h, says idiot openbsd coder */ |
65 |
|
|
#if HAVE_STDINT_H |
66 |
|
|
# include <stdint.h> |
67 |
|
|
#endif |
68 |
|
|
|
69 |
root |
1.1 |
#ifndef EIO_FINISH |
70 |
|
|
# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
71 |
|
|
#endif |
72 |
|
|
|
73 |
|
|
#ifndef EIO_DESTROY |
74 |
|
|
# define EIO_DESTROY(req) do { if ((req)->destroy) (req)->destroy (req); } while (0) |
75 |
|
|
#endif |
76 |
|
|
|
77 |
|
|
#ifndef EIO_FEED |
78 |
|
|
# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
79 |
|
|
#endif |
80 |
|
|
|
81 |
|
|
#ifdef _WIN32 |
82 |
|
|
|
83 |
|
|
/*doh*/ |
84 |
|
|
#else |
85 |
|
|
|
86 |
|
|
# include <sys/time.h> |
87 |
|
|
# include <sys/select.h> |
88 |
|
|
# include <unistd.h> |
89 |
|
|
# include <utime.h> |
90 |
|
|
# include <signal.h> |
91 |
root |
1.4 |
# include <dirent.h> |
92 |
root |
1.1 |
|
93 |
root |
1.60 |
#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
94 |
root |
1.56 |
# include <sys/mman.h> |
95 |
|
|
#endif |
96 |
|
|
|
97 |
root |
1.31 |
/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
98 |
root |
1.50 |
# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
99 |
root |
1.33 |
# define _DIRENT_HAVE_D_TYPE /* sigh */ |
100 |
root |
1.31 |
# define D_INO(de) (de)->d_fileno |
101 |
root |
1.33 |
# define D_NAMLEN(de) (de)->d_namlen |
102 |
root |
1.45 |
# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
103 |
root |
1.31 |
# define D_INO(de) (de)->d_ino |
104 |
|
|
# endif |
105 |
|
|
|
106 |
root |
1.33 |
#ifdef _D_EXACT_NAMLEN |
107 |
|
|
# undef D_NAMLEN |
108 |
|
|
# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
109 |
|
|
#endif |
110 |
|
|
|
111 |
root |
1.31 |
# ifdef _DIRENT_HAVE_D_TYPE |
112 |
|
|
# define D_TYPE(de) (de)->d_type |
113 |
|
|
# endif |
114 |
|
|
|
115 |
root |
1.1 |
# ifndef EIO_STRUCT_DIRENT |
116 |
|
|
# define EIO_STRUCT_DIRENT struct dirent |
117 |
|
|
# endif |
118 |
|
|
|
119 |
|
|
#endif |
120 |
|
|
|
121 |
|
|
#if HAVE_SENDFILE |
122 |
|
|
# if __linux |
123 |
|
|
# include <sys/sendfile.h> |
124 |
root |
1.50 |
# elif __FreeBSD__ || defined __APPLE__ |
125 |
root |
1.1 |
# include <sys/socket.h> |
126 |
|
|
# include <sys/uio.h> |
127 |
|
|
# elif __hpux |
128 |
|
|
# include <sys/socket.h> |
129 |
root |
1.45 |
# elif __solaris |
130 |
root |
1.1 |
# include <sys/sendfile.h> |
131 |
|
|
# else |
132 |
|
|
# error sendfile support requested but not available |
133 |
|
|
# endif |
134 |
|
|
#endif |
135 |
|
|
|
136 |
root |
1.31 |
#ifndef D_TYPE |
137 |
|
|
# define D_TYPE(de) 0 |
138 |
|
|
#endif |
139 |
|
|
#ifndef D_INO |
140 |
|
|
# define D_INO(de) 0 |
141 |
|
|
#endif |
142 |
root |
1.33 |
#ifndef D_NAMLEN |
143 |
|
|
# define D_NAMLEN(de) strlen ((de)->d_name) |
144 |
|
|
#endif |
145 |
root |
1.31 |
|
146 |
root |
1.1 |
/* used for struct dirent, AIX doesn't provide it */ |
147 |
|
|
#ifndef NAME_MAX |
148 |
|
|
# define NAME_MAX 4096 |
149 |
|
|
#endif |
150 |
|
|
|
151 |
root |
1.54 |
/* used for readlink etc. */ |
152 |
|
|
#ifndef PATH_MAX |
153 |
|
|
# define PATH_MAX 4096 |
154 |
|
|
#endif |
155 |
|
|
|
156 |
root |
1.1 |
/* buffer size for various temporary buffers */ |
157 |
|
|
#define EIO_BUFSIZE 65536 |
158 |
|
|
|
159 |
|
|
#define dBUF \ |
160 |
|
|
char *eio_buf; \ |
161 |
root |
1.14 |
ETP_WORKER_LOCK (self); \ |
162 |
root |
1.1 |
self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
163 |
root |
1.14 |
ETP_WORKER_UNLOCK (self); \ |
164 |
root |
1.9 |
errno = ENOMEM; \ |
165 |
root |
1.1 |
if (!eio_buf) \ |
166 |
|
|
return -1; |
167 |
|
|
|
168 |
|
|
#define EIO_TICKS ((1000000 + 1023) >> 10) |
169 |
|
|
|
170 |
root |
1.13 |
/*****************************************************************************/ |
171 |
root |
1.1 |
|
172 |
root |
1.24 |
#if __GNUC__ >= 3 |
173 |
|
|
# define expect(expr,value) __builtin_expect ((expr),(value)) |
174 |
|
|
#else |
175 |
|
|
# define expect(expr,value) (expr) |
176 |
|
|
#endif |
177 |
|
|
|
178 |
|
|
#define expect_false(expr) expect ((expr) != 0, 0) |
179 |
|
|
#define expect_true(expr) expect ((expr) != 0, 1) |
180 |
|
|
|
181 |
|
|
/*****************************************************************************/ |
182 |
|
|
|
183 |
root |
1.14 |
#define ETP_PRI_MIN EIO_PRI_MIN |
184 |
|
|
#define ETP_PRI_MAX EIO_PRI_MAX |
185 |
|
|
|
186 |
root |
1.15 |
struct etp_worker; |
187 |
|
|
|
188 |
root |
1.14 |
#define ETP_REQ eio_req |
189 |
|
|
#define ETP_DESTROY(req) eio_destroy (req) |
190 |
|
|
static int eio_finish (eio_req *req); |
191 |
|
|
#define ETP_FINISH(req) eio_finish (req) |
192 |
root |
1.15 |
static void eio_execute (struct etp_worker *self, eio_req *req); |
193 |
|
|
#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
194 |
root |
1.14 |
|
195 |
|
|
#define ETP_WORKER_CLEAR(req) \ |
196 |
|
|
if (wrk->dbuf) \ |
197 |
|
|
{ \ |
198 |
|
|
free (wrk->dbuf); \ |
199 |
|
|
wrk->dbuf = 0; \ |
200 |
|
|
} \ |
201 |
|
|
\ |
202 |
|
|
if (wrk->dirp) \ |
203 |
|
|
{ \ |
204 |
|
|
closedir (wrk->dirp); \ |
205 |
|
|
wrk->dirp = 0; \ |
206 |
|
|
} |
207 |
root |
1.32 |
|
208 |
root |
1.14 |
#define ETP_WORKER_COMMON \ |
209 |
|
|
void *dbuf; \ |
210 |
|
|
DIR *dirp; |
211 |
|
|
|
212 |
|
|
/*****************************************************************************/ |
213 |
|
|
|
214 |
|
|
#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
215 |
|
|
|
216 |
root |
1.44 |
/* calculate time difference in ~1/EIO_TICKS of a second */ |
217 |
root |
1.1 |
static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
218 |
|
|
{ |
219 |
|
|
return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
220 |
|
|
+ ((tv2->tv_usec - tv1->tv_usec) >> 10); |
221 |
|
|
} |
222 |
|
|
|
223 |
root |
1.9 |
static unsigned int started, idle, wanted = 4; |
224 |
root |
1.1 |
|
225 |
root |
1.14 |
static void (*want_poll_cb) (void); |
226 |
|
|
static void (*done_poll_cb) (void); |
227 |
root |
1.13 |
|
228 |
root |
1.14 |
static unsigned int max_poll_time; /* reslock */ |
229 |
|
|
static unsigned int max_poll_reqs; /* reslock */ |
230 |
root |
1.13 |
|
231 |
root |
1.14 |
static volatile unsigned int nreqs; /* reqlock */ |
232 |
|
|
static volatile unsigned int nready; /* reqlock */ |
233 |
|
|
static volatile unsigned int npending; /* reqlock */ |
234 |
root |
1.62 |
static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
235 |
|
|
static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
236 |
root |
1.13 |
|
237 |
root |
1.62 |
static xmutex_t wrklock; |
238 |
|
|
static xmutex_t reslock; |
239 |
|
|
static xmutex_t reqlock; |
240 |
|
|
static xcond_t reqwait; |
241 |
root |
1.1 |
|
242 |
root |
1.22 |
#if !HAVE_PREADWRITE |
243 |
|
|
/* |
244 |
|
|
* make our pread/pwrite emulation safe against themselves, but not against |
245 |
|
|
* normal read/write by using a mutex. slows down execution a lot, |
246 |
|
|
* but that's your problem, not mine. |
247 |
|
|
*/ |
248 |
root |
1.55 |
static xmutex_t preadwritelock = X_MUTEX_INIT; |
249 |
root |
1.22 |
#endif |
250 |
|
|
|
251 |
root |
1.14 |
typedef struct etp_worker |
252 |
root |
1.9 |
{ |
253 |
root |
1.14 |
/* locked by wrklock */ |
254 |
|
|
struct etp_worker *prev, *next; |
255 |
root |
1.1 |
|
256 |
root |
1.55 |
xthread_t tid; |
257 |
root |
1.1 |
|
258 |
root |
1.14 |
/* locked by reslock, reqlock or wrklock */ |
259 |
|
|
ETP_REQ *req; /* currently processed request */ |
260 |
|
|
|
261 |
|
|
ETP_WORKER_COMMON |
262 |
|
|
} etp_worker; |
263 |
|
|
|
264 |
|
|
static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
265 |
root |
1.1 |
|
266 |
root |
1.14 |
#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
267 |
|
|
#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
268 |
root |
1.13 |
|
269 |
|
|
/* worker threads management */ |
270 |
root |
1.1 |
|
271 |
root |
1.14 |
static void etp_worker_clear (etp_worker *wrk) |
272 |
root |
1.1 |
{ |
273 |
root |
1.14 |
ETP_WORKER_CLEAR (wrk); |
274 |
root |
1.1 |
} |
275 |
|
|
|
276 |
root |
1.14 |
static void etp_worker_free (etp_worker *wrk) |
277 |
root |
1.1 |
{ |
278 |
|
|
wrk->next->prev = wrk->prev; |
279 |
|
|
wrk->prev->next = wrk->next; |
280 |
|
|
|
281 |
|
|
free (wrk); |
282 |
|
|
} |
283 |
|
|
|
284 |
root |
1.14 |
static unsigned int etp_nreqs (void) |
285 |
root |
1.1 |
{ |
286 |
root |
1.14 |
int retval; |
287 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
288 |
|
|
retval = nreqs; |
289 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
290 |
|
|
return retval; |
291 |
root |
1.1 |
} |
292 |
|
|
|
293 |
root |
1.14 |
static unsigned int etp_nready (void) |
294 |
root |
1.1 |
{ |
295 |
|
|
unsigned int retval; |
296 |
|
|
|
297 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
298 |
|
|
retval = nready; |
299 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
300 |
|
|
|
301 |
|
|
return retval; |
302 |
|
|
} |
303 |
|
|
|
304 |
root |
1.14 |
static unsigned int etp_npending (void) |
305 |
root |
1.1 |
{ |
306 |
|
|
unsigned int retval; |
307 |
|
|
|
308 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
309 |
|
|
retval = npending; |
310 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
311 |
|
|
|
312 |
|
|
return retval; |
313 |
|
|
} |
314 |
|
|
|
315 |
root |
1.14 |
static unsigned int etp_nthreads (void) |
316 |
root |
1.1 |
{ |
317 |
|
|
unsigned int retval; |
318 |
|
|
|
319 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
320 |
|
|
retval = started; |
321 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
322 |
|
|
|
323 |
|
|
return retval; |
324 |
|
|
} |
325 |
|
|
|
326 |
|
|
/* |
327 |
|
|
* a somewhat faster data structure might be nice, but |
328 |
|
|
* with 8 priorities this actually needs <20 insns |
329 |
|
|
* per shift, the most expensive operation. |
330 |
|
|
*/ |
331 |
|
|
typedef struct { |
332 |
root |
1.14 |
ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */ |
333 |
root |
1.1 |
int size; |
334 |
root |
1.14 |
} etp_reqq; |
335 |
root |
1.1 |
|
336 |
root |
1.14 |
static etp_reqq req_queue; |
337 |
|
|
static etp_reqq res_queue; |
338 |
root |
1.1 |
|
339 |
root |
1.14 |
static int reqq_push (etp_reqq *q, ETP_REQ *req) |
340 |
root |
1.1 |
{ |
341 |
|
|
int pri = req->pri; |
342 |
|
|
req->next = 0; |
343 |
|
|
|
344 |
|
|
if (q->qe[pri]) |
345 |
|
|
{ |
346 |
|
|
q->qe[pri]->next = req; |
347 |
|
|
q->qe[pri] = req; |
348 |
|
|
} |
349 |
|
|
else |
350 |
|
|
q->qe[pri] = q->qs[pri] = req; |
351 |
|
|
|
352 |
|
|
return q->size++; |
353 |
|
|
} |
354 |
|
|
|
355 |
root |
1.14 |
static ETP_REQ *reqq_shift (etp_reqq *q) |
356 |
root |
1.1 |
{ |
357 |
|
|
int pri; |
358 |
|
|
|
359 |
|
|
if (!q->size) |
360 |
|
|
return 0; |
361 |
|
|
|
362 |
|
|
--q->size; |
363 |
|
|
|
364 |
root |
1.14 |
for (pri = ETP_NUM_PRI; pri--; ) |
365 |
root |
1.1 |
{ |
366 |
|
|
eio_req *req = q->qs[pri]; |
367 |
|
|
|
368 |
|
|
if (req) |
369 |
|
|
{ |
370 |
|
|
if (!(q->qs[pri] = (eio_req *)req->next)) |
371 |
|
|
q->qe[pri] = 0; |
372 |
|
|
|
373 |
|
|
return req; |
374 |
|
|
} |
375 |
|
|
} |
376 |
|
|
|
377 |
|
|
abort (); |
378 |
|
|
} |
379 |
|
|
|
380 |
root |
1.62 |
static void etp_thread_init (void) |
381 |
|
|
{ |
382 |
|
|
X_MUTEX_CREATE (wrklock); |
383 |
|
|
X_MUTEX_CREATE (reslock); |
384 |
|
|
X_MUTEX_CREATE (reqlock); |
385 |
|
|
X_COND_CREATE (reqwait); |
386 |
|
|
} |
387 |
|
|
|
388 |
root |
1.13 |
static void etp_atfork_prepare (void) |
389 |
|
|
{ |
390 |
root |
1.14 |
X_LOCK (wrklock); |
391 |
root |
1.13 |
X_LOCK (reqlock); |
392 |
|
|
X_LOCK (reslock); |
393 |
|
|
#if !HAVE_PREADWRITE |
394 |
|
|
X_LOCK (preadwritelock); |
395 |
|
|
#endif |
396 |
|
|
} |
397 |
|
|
|
398 |
|
|
static void etp_atfork_parent (void) |
399 |
|
|
{ |
400 |
|
|
#if !HAVE_PREADWRITE |
401 |
|
|
X_UNLOCK (preadwritelock); |
402 |
|
|
#endif |
403 |
|
|
X_UNLOCK (reslock); |
404 |
|
|
X_UNLOCK (reqlock); |
405 |
root |
1.14 |
X_UNLOCK (wrklock); |
406 |
root |
1.13 |
} |
407 |
|
|
|
408 |
|
|
static void etp_atfork_child (void) |
409 |
|
|
{ |
410 |
root |
1.14 |
ETP_REQ *prv; |
411 |
root |
1.13 |
|
412 |
root |
1.21 |
while ((prv = reqq_shift (&req_queue))) |
413 |
root |
1.14 |
ETP_DESTROY (prv); |
414 |
root |
1.13 |
|
415 |
root |
1.21 |
while ((prv = reqq_shift (&res_queue))) |
416 |
root |
1.14 |
ETP_DESTROY (prv); |
417 |
root |
1.13 |
|
418 |
|
|
while (wrk_first.next != &wrk_first) |
419 |
|
|
{ |
420 |
root |
1.14 |
etp_worker *wrk = wrk_first.next; |
421 |
root |
1.13 |
|
422 |
|
|
if (wrk->req) |
423 |
root |
1.14 |
ETP_DESTROY (wrk->req); |
424 |
root |
1.13 |
|
425 |
root |
1.14 |
etp_worker_clear (wrk); |
426 |
|
|
etp_worker_free (wrk); |
427 |
root |
1.13 |
} |
428 |
|
|
|
429 |
|
|
started = 0; |
430 |
|
|
idle = 0; |
431 |
|
|
nreqs = 0; |
432 |
|
|
nready = 0; |
433 |
|
|
npending = 0; |
434 |
|
|
|
435 |
root |
1.62 |
etp_thread_init (); |
436 |
root |
1.13 |
} |
437 |
|
|
|
438 |
|
|
static void |
439 |
|
|
etp_once_init (void) |
440 |
root |
1.62 |
{ |
441 |
|
|
etp_thread_init (); |
442 |
root |
1.13 |
X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
443 |
|
|
} |
444 |
|
|
|
445 |
|
|
static int |
446 |
root |
1.14 |
etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
447 |
root |
1.13 |
{ |
448 |
|
|
static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
449 |
|
|
|
450 |
|
|
pthread_once (&doinit, etp_once_init); |
451 |
|
|
|
452 |
root |
1.14 |
want_poll_cb = want_poll; |
453 |
|
|
done_poll_cb = done_poll; |
454 |
root |
1.19 |
|
455 |
|
|
return 0; |
456 |
root |
1.14 |
} |
457 |
|
|
|
458 |
|
|
X_THREAD_PROC (etp_proc); |
459 |
|
|
|
460 |
|
|
static void etp_start_thread (void) |
461 |
|
|
{ |
462 |
|
|
etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
463 |
|
|
|
464 |
|
|
/*TODO*/ |
465 |
|
|
assert (("unable to allocate worker thread data", wrk)); |
466 |
|
|
|
467 |
|
|
X_LOCK (wrklock); |
468 |
|
|
|
469 |
|
|
if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) |
470 |
|
|
{ |
471 |
|
|
wrk->prev = &wrk_first; |
472 |
|
|
wrk->next = wrk_first.next; |
473 |
|
|
wrk_first.next->prev = wrk; |
474 |
|
|
wrk_first.next = wrk; |
475 |
|
|
++started; |
476 |
|
|
} |
477 |
|
|
else |
478 |
|
|
free (wrk); |
479 |
|
|
|
480 |
|
|
X_UNLOCK (wrklock); |
481 |
|
|
} |
482 |
|
|
|
483 |
|
|
static void etp_maybe_start_thread (void) |
484 |
|
|
{ |
485 |
root |
1.24 |
if (expect_true (etp_nthreads () >= wanted)) |
486 |
root |
1.14 |
return; |
487 |
|
|
|
488 |
|
|
/* todo: maybe use idle here, but might be less exact */ |
489 |
root |
1.24 |
if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
490 |
root |
1.14 |
return; |
491 |
|
|
|
492 |
|
|
etp_start_thread (); |
493 |
|
|
} |
494 |
|
|
|
495 |
|
|
static void etp_end_thread (void) |
496 |
|
|
{ |
497 |
|
|
eio_req *req = calloc (1, sizeof (eio_req)); |
498 |
|
|
|
499 |
|
|
req->type = -1; |
500 |
|
|
req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
501 |
|
|
|
502 |
|
|
X_LOCK (reqlock); |
503 |
|
|
reqq_push (&req_queue, req); |
504 |
|
|
X_COND_SIGNAL (reqwait); |
505 |
|
|
X_UNLOCK (reqlock); |
506 |
|
|
|
507 |
|
|
X_LOCK (wrklock); |
508 |
|
|
--started; |
509 |
|
|
X_UNLOCK (wrklock); |
510 |
|
|
} |
511 |
|
|
|
512 |
|
|
static int etp_poll (void) |
513 |
|
|
{ |
514 |
|
|
unsigned int maxreqs; |
515 |
|
|
unsigned int maxtime; |
516 |
|
|
struct timeval tv_start, tv_now; |
517 |
|
|
|
518 |
|
|
X_LOCK (reslock); |
519 |
|
|
maxreqs = max_poll_reqs; |
520 |
|
|
maxtime = max_poll_time; |
521 |
|
|
X_UNLOCK (reslock); |
522 |
|
|
|
523 |
|
|
if (maxtime) |
524 |
|
|
gettimeofday (&tv_start, 0); |
525 |
|
|
|
526 |
|
|
for (;;) |
527 |
|
|
{ |
528 |
|
|
ETP_REQ *req; |
529 |
|
|
|
530 |
|
|
etp_maybe_start_thread (); |
531 |
|
|
|
532 |
|
|
X_LOCK (reslock); |
533 |
|
|
req = reqq_shift (&res_queue); |
534 |
|
|
|
535 |
|
|
if (req) |
536 |
|
|
{ |
537 |
|
|
--npending; |
538 |
|
|
|
539 |
|
|
if (!res_queue.size && done_poll_cb) |
540 |
|
|
done_poll_cb (); |
541 |
|
|
} |
542 |
|
|
|
543 |
|
|
X_UNLOCK (reslock); |
544 |
|
|
|
545 |
|
|
if (!req) |
546 |
|
|
return 0; |
547 |
|
|
|
548 |
|
|
X_LOCK (reqlock); |
549 |
|
|
--nreqs; |
550 |
|
|
X_UNLOCK (reqlock); |
551 |
|
|
|
552 |
root |
1.24 |
if (expect_false (req->type == EIO_GROUP && req->size)) |
553 |
root |
1.14 |
{ |
554 |
|
|
req->int1 = 1; /* mark request as delayed */ |
555 |
|
|
continue; |
556 |
|
|
} |
557 |
|
|
else |
558 |
|
|
{ |
559 |
|
|
int res = ETP_FINISH (req); |
560 |
root |
1.24 |
if (expect_false (res)) |
561 |
root |
1.14 |
return res; |
562 |
|
|
} |
563 |
|
|
|
564 |
root |
1.24 |
if (expect_false (maxreqs && !--maxreqs)) |
565 |
root |
1.14 |
break; |
566 |
|
|
|
567 |
|
|
if (maxtime) |
568 |
|
|
{ |
569 |
|
|
gettimeofday (&tv_now, 0); |
570 |
|
|
|
571 |
|
|
if (tvdiff (&tv_start, &tv_now) >= maxtime) |
572 |
|
|
break; |
573 |
|
|
} |
574 |
|
|
} |
575 |
|
|
|
576 |
|
|
errno = EAGAIN; |
577 |
|
|
return -1; |
578 |
|
|
} |
579 |
|
|
|
580 |
|
|
static void etp_cancel (ETP_REQ *req) |
581 |
|
|
{ |
582 |
|
|
X_LOCK (wrklock); |
583 |
|
|
req->flags |= EIO_FLAG_CANCELLED; |
584 |
|
|
X_UNLOCK (wrklock); |
585 |
|
|
|
586 |
|
|
eio_grp_cancel (req); |
587 |
|
|
} |
588 |
|
|
|
589 |
|
|
static void etp_submit (ETP_REQ *req) |
590 |
|
|
{ |
591 |
|
|
req->pri -= ETP_PRI_MIN; |
592 |
|
|
|
593 |
root |
1.24 |
if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
594 |
|
|
if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
595 |
|
|
|
596 |
|
|
if (expect_false (req->type == EIO_GROUP)) |
597 |
|
|
{ |
598 |
|
|
/* I hope this is worth it :/ */ |
599 |
|
|
X_LOCK (reqlock); |
600 |
|
|
++nreqs; |
601 |
|
|
X_UNLOCK (reqlock); |
602 |
|
|
|
603 |
|
|
X_LOCK (reslock); |
604 |
|
|
|
605 |
|
|
++npending; |
606 |
|
|
|
607 |
|
|
if (!reqq_push (&res_queue, req) && want_poll_cb) |
608 |
|
|
want_poll_cb (); |
609 |
root |
1.14 |
|
610 |
root |
1.24 |
X_UNLOCK (reslock); |
611 |
|
|
} |
612 |
|
|
else |
613 |
|
|
{ |
614 |
|
|
X_LOCK (reqlock); |
615 |
|
|
++nreqs; |
616 |
|
|
++nready; |
617 |
|
|
reqq_push (&req_queue, req); |
618 |
|
|
X_COND_SIGNAL (reqwait); |
619 |
|
|
X_UNLOCK (reqlock); |
620 |
root |
1.14 |
|
621 |
root |
1.24 |
etp_maybe_start_thread (); |
622 |
|
|
} |
623 |
root |
1.14 |
} |
624 |
|
|
|
625 |
|
|
static void etp_set_max_poll_time (double nseconds) |
626 |
|
|
{ |
627 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
628 |
root |
1.44 |
max_poll_time = nseconds * EIO_TICKS; |
629 |
root |
1.14 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
630 |
|
|
} |
631 |
|
|
|
632 |
|
|
static void etp_set_max_poll_reqs (unsigned int maxreqs) |
633 |
|
|
{ |
634 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
635 |
|
|
max_poll_reqs = maxreqs; |
636 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
637 |
|
|
} |
638 |
root |
1.13 |
|
639 |
root |
1.14 |
static void etp_set_max_idle (unsigned int nthreads) |
640 |
|
|
{ |
641 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
642 |
root |
1.62 |
max_idle = nthreads; |
643 |
|
|
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
644 |
|
|
} |
645 |
|
|
|
646 |
|
|
static void etp_set_idle_timeout (unsigned int seconds) |
647 |
|
|
{ |
648 |
|
|
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
649 |
|
|
idle_timeout = seconds; |
650 |
root |
1.14 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
651 |
root |
1.13 |
} |
652 |
|
|
|
653 |
root |
1.14 |
static void etp_set_min_parallel (unsigned int nthreads) |
654 |
|
|
{ |
655 |
|
|
if (wanted < nthreads) |
656 |
|
|
wanted = nthreads; |
657 |
|
|
} |
658 |
|
|
|
659 |
|
|
static void etp_set_max_parallel (unsigned int nthreads) |
660 |
|
|
{ |
661 |
|
|
if (wanted > nthreads) |
662 |
|
|
wanted = nthreads; |
663 |
|
|
|
664 |
|
|
while (started > wanted) |
665 |
|
|
etp_end_thread (); |
666 |
|
|
} |
667 |
root |
1.13 |
|
668 |
|
|
/*****************************************************************************/ |
669 |
|
|
|
670 |
root |
1.2 |
static void grp_try_feed (eio_req *grp) |
671 |
root |
1.1 |
{ |
672 |
|
|
while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
673 |
|
|
{ |
674 |
root |
1.26 |
grp->flags &= ~EIO_FLAG_GROUPADD; |
675 |
root |
1.1 |
|
676 |
|
|
EIO_FEED (grp); |
677 |
|
|
|
678 |
|
|
/* stop if no progress has been made */ |
679 |
root |
1.26 |
if (!(grp->flags & EIO_FLAG_GROUPADD)) |
680 |
root |
1.1 |
{ |
681 |
|
|
grp->feed = 0; |
682 |
root |
1.2 |
break; |
683 |
root |
1.1 |
} |
684 |
|
|
} |
685 |
|
|
} |
686 |
|
|
|
687 |
|
|
static int grp_dec (eio_req *grp) |
688 |
|
|
{ |
689 |
|
|
--grp->size; |
690 |
|
|
|
691 |
|
|
/* call feeder, if applicable */ |
692 |
root |
1.2 |
grp_try_feed (grp); |
693 |
root |
1.1 |
|
694 |
|
|
/* finish, if done */ |
695 |
|
|
if (!grp->size && grp->int1) |
696 |
root |
1.2 |
return eio_finish (grp); |
697 |
root |
1.1 |
else |
698 |
|
|
return 0; |
699 |
|
|
} |
700 |
|
|
|
701 |
|
|
void eio_destroy (eio_req *req) |
702 |
|
|
{ |
703 |
root |
1.6 |
if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
704 |
|
|
if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
705 |
root |
1.1 |
|
706 |
|
|
EIO_DESTROY (req); |
707 |
|
|
} |
708 |
|
|
|
709 |
root |
1.2 |
static int eio_finish (eio_req *req) |
710 |
root |
1.1 |
{ |
711 |
|
|
int res = EIO_FINISH (req); |
712 |
|
|
|
713 |
|
|
if (req->grp) |
714 |
|
|
{ |
715 |
|
|
int res2; |
716 |
|
|
eio_req *grp = req->grp; |
717 |
|
|
|
718 |
|
|
/* unlink request */ |
719 |
|
|
if (req->grp_next) req->grp_next->grp_prev = req->grp_prev; |
720 |
|
|
if (req->grp_prev) req->grp_prev->grp_next = req->grp_next; |
721 |
|
|
|
722 |
|
|
if (grp->grp_first == req) |
723 |
|
|
grp->grp_first = req->grp_next; |
724 |
|
|
|
725 |
|
|
res2 = grp_dec (grp); |
726 |
|
|
|
727 |
|
|
if (!res && res2) |
728 |
|
|
res = res2; |
729 |
|
|
} |
730 |
|
|
|
731 |
|
|
eio_destroy (req); |
732 |
|
|
|
733 |
|
|
return res; |
734 |
|
|
} |
735 |
|
|
|
736 |
|
|
void eio_grp_cancel (eio_req *grp) |
737 |
|
|
{ |
738 |
|
|
for (grp = grp->grp_first; grp; grp = grp->grp_next) |
739 |
|
|
eio_cancel (grp); |
740 |
|
|
} |
741 |
|
|
|
742 |
|
|
void eio_cancel (eio_req *req) |
743 |
|
|
{ |
744 |
root |
1.14 |
etp_cancel (req); |
745 |
|
|
} |
746 |
root |
1.1 |
|
747 |
root |
1.14 |
void eio_submit (eio_req *req) |
748 |
|
|
{ |
749 |
|
|
etp_submit (req); |
750 |
root |
1.1 |
} |
751 |
|
|
|
752 |
root |
1.14 |
unsigned int eio_nreqs (void) |
753 |
root |
1.1 |
{ |
754 |
root |
1.14 |
return etp_nreqs (); |
755 |
root |
1.1 |
} |
756 |
|
|
|
757 |
root |
1.14 |
unsigned int eio_nready (void) |
758 |
root |
1.1 |
{ |
759 |
root |
1.14 |
return etp_nready (); |
760 |
root |
1.1 |
} |
761 |
|
|
|
762 |
root |
1.14 |
unsigned int eio_npending (void) |
763 |
root |
1.1 |
{ |
764 |
root |
1.14 |
return etp_npending (); |
765 |
root |
1.1 |
} |
766 |
|
|
|
767 |
root |
1.14 |
unsigned int eio_nthreads (void) |
768 |
root |
1.1 |
{ |
769 |
root |
1.14 |
return etp_nthreads (); |
770 |
root |
1.1 |
} |
771 |
|
|
|
772 |
|
|
void eio_set_max_poll_time (double nseconds) |
773 |
|
|
{ |
774 |
root |
1.14 |
etp_set_max_poll_time (nseconds); |
775 |
root |
1.1 |
} |
776 |
|
|
|
777 |
|
|
void eio_set_max_poll_reqs (unsigned int maxreqs) |
778 |
|
|
{ |
779 |
root |
1.14 |
etp_set_max_poll_reqs (maxreqs); |
780 |
root |
1.1 |
} |
781 |
|
|
|
782 |
|
|
void eio_set_max_idle (unsigned int nthreads) |
783 |
|
|
{ |
784 |
root |
1.14 |
etp_set_max_idle (nthreads); |
785 |
root |
1.1 |
} |
786 |
|
|
|
787 |
root |
1.62 |
void eio_set_idle_timeout (unsigned int seconds) |
788 |
|
|
{ |
789 |
|
|
etp_set_idle_timeout (seconds); |
790 |
|
|
} |
791 |
|
|
|
792 |
root |
1.1 |
void eio_set_min_parallel (unsigned int nthreads) |
793 |
|
|
{ |
794 |
root |
1.14 |
etp_set_min_parallel (nthreads); |
795 |
root |
1.1 |
} |
796 |
|
|
|
797 |
|
|
void eio_set_max_parallel (unsigned int nthreads) |
798 |
|
|
{ |
799 |
root |
1.14 |
etp_set_max_parallel (nthreads); |
800 |
root |
1.1 |
} |
801 |
|
|
|
802 |
|
|
int eio_poll (void) |
803 |
|
|
{ |
804 |
root |
1.14 |
return etp_poll (); |
805 |
root |
1.1 |
} |
806 |
|
|
|
807 |
|
|
/*****************************************************************************/ |
808 |
|
|
/* work around various missing functions */ |
809 |
|
|
|
810 |
|
|
#if !HAVE_PREADWRITE |
811 |
root |
1.27 |
# undef pread |
812 |
|
|
# undef pwrite |
813 |
root |
1.9 |
# define pread eio__pread |
814 |
|
|
# define pwrite eio__pwrite |
815 |
root |
1.1 |
|
816 |
root |
1.9 |
static ssize_t |
817 |
|
|
eio__pread (int fd, void *buf, size_t count, off_t offset) |
818 |
root |
1.1 |
{ |
819 |
|
|
ssize_t res; |
820 |
|
|
off_t ooffset; |
821 |
|
|
|
822 |
|
|
X_LOCK (preadwritelock); |
823 |
|
|
ooffset = lseek (fd, 0, SEEK_CUR); |
824 |
|
|
lseek (fd, offset, SEEK_SET); |
825 |
|
|
res = read (fd, buf, count); |
826 |
|
|
lseek (fd, ooffset, SEEK_SET); |
827 |
|
|
X_UNLOCK (preadwritelock); |
828 |
|
|
|
829 |
|
|
return res; |
830 |
|
|
} |
831 |
|
|
|
832 |
root |
1.9 |
static ssize_t |
833 |
|
|
eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
834 |
root |
1.1 |
{ |
835 |
|
|
ssize_t res; |
836 |
|
|
off_t ooffset; |
837 |
|
|
|
838 |
|
|
X_LOCK (preadwritelock); |
839 |
|
|
ooffset = lseek (fd, 0, SEEK_CUR); |
840 |
|
|
lseek (fd, offset, SEEK_SET); |
841 |
|
|
res = write (fd, buf, count); |
842 |
root |
1.30 |
lseek (fd, ooffset, SEEK_SET); |
843 |
root |
1.1 |
X_UNLOCK (preadwritelock); |
844 |
|
|
|
845 |
|
|
return res; |
846 |
|
|
} |
847 |
|
|
#endif |
848 |
|
|
|
849 |
root |
1.59 |
#ifndef HAVE_UTIMES |
850 |
root |
1.1 |
|
851 |
root |
1.27 |
# undef utimes |
852 |
root |
1.59 |
# define utimes(path,times) eio__utimes (path, times) |
853 |
root |
1.1 |
|
854 |
root |
1.9 |
static int |
855 |
|
|
eio__utimes (const char *filename, const struct timeval times[2]) |
856 |
root |
1.1 |
{ |
857 |
|
|
if (times) |
858 |
|
|
{ |
859 |
|
|
struct utimbuf buf; |
860 |
|
|
|
861 |
|
|
buf.actime = times[0].tv_sec; |
862 |
|
|
buf.modtime = times[1].tv_sec; |
863 |
|
|
|
864 |
|
|
return utime (filename, &buf); |
865 |
|
|
} |
866 |
|
|
else |
867 |
|
|
return utime (filename, 0); |
868 |
|
|
} |
869 |
|
|
|
870 |
root |
1.59 |
#endif |
871 |
|
|
|
872 |
|
|
#ifndef HAVE_FUTIMES |
873 |
|
|
|
874 |
|
|
# undef futimes |
875 |
|
|
# define futimes(fd,times) eio__futimes (fd, times) |
876 |
|
|
|
877 |
root |
1.9 |
static int eio__futimes (int fd, const struct timeval tv[2]) |
878 |
root |
1.1 |
{ |
879 |
|
|
errno = ENOSYS; |
880 |
|
|
return -1; |
881 |
|
|
} |
882 |
|
|
|
883 |
|
|
#endif |
884 |
|
|
|
885 |
|
|
#if !HAVE_FDATASYNC |
886 |
root |
1.27 |
# undef fdatasync |
887 |
|
|
# define fdatasync(fd) fsync (fd) |
888 |
root |
1.1 |
#endif |
889 |
|
|
|
890 |
root |
1.27 |
/* sync_file_range always needs emulation */ |
891 |
|
|
int |
892 |
|
|
eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
893 |
|
|
{ |
894 |
|
|
#if HAVE_SYNC_FILE_RANGE |
895 |
|
|
int res; |
896 |
|
|
|
897 |
|
|
if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
898 |
|
|
|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
899 |
|
|
|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
900 |
|
|
{ |
901 |
|
|
flags = 0 |
902 |
|
|
| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
903 |
|
|
| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
904 |
|
|
| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
905 |
|
|
} |
906 |
|
|
|
907 |
|
|
res = sync_file_range (fd, offset, nbytes, flags); |
908 |
|
|
|
909 |
root |
1.28 |
if (!res || errno != ENOSYS) |
910 |
root |
1.27 |
return res; |
911 |
|
|
#endif |
912 |
|
|
|
913 |
|
|
/* even though we could play tricks with the flags, it's better to always |
914 |
root |
1.43 |
* call fdatasync, as that matches the expectation of its users best */ |
915 |
root |
1.27 |
return fdatasync (fd); |
916 |
|
|
} |
917 |
|
|
|
918 |
root |
1.1 |
#if !HAVE_READAHEAD |
919 |
root |
1.27 |
# undef readahead |
920 |
root |
1.9 |
# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
921 |
root |
1.1 |
|
922 |
root |
1.9 |
static ssize_t |
923 |
root |
1.17 |
eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
924 |
root |
1.1 |
{ |
925 |
|
|
size_t todo = count; |
926 |
|
|
dBUF; |
927 |
|
|
|
928 |
|
|
while (todo > 0) |
929 |
|
|
{ |
930 |
|
|
size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; |
931 |
|
|
|
932 |
root |
1.3 |
pread (fd, eio_buf, len, offset); |
933 |
root |
1.1 |
offset += len; |
934 |
|
|
todo -= len; |
935 |
|
|
} |
936 |
|
|
|
937 |
|
|
errno = 0; |
938 |
|
|
return count; |
939 |
|
|
} |
940 |
|
|
|
941 |
|
|
#endif |
942 |
|
|
|
943 |
|
|
/* sendfile always needs emulation */ |
944 |
root |
1.9 |
static ssize_t |
945 |
root |
1.14 |
eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
946 |
root |
1.1 |
{ |
947 |
root |
1.69 |
ssize_t written = 0; |
948 |
root |
1.1 |
ssize_t res; |
949 |
|
|
|
950 |
|
|
if (!count) |
951 |
|
|
return 0; |
952 |
|
|
|
953 |
root |
1.69 |
for (;;) |
954 |
|
|
{ |
955 |
root |
1.1 |
#if HAVE_SENDFILE |
956 |
|
|
# if __linux |
957 |
root |
1.69 |
off_t soffset = offset; |
958 |
|
|
res = sendfile (ofd, ifd, &soffset, count); |
959 |
root |
1.1 |
|
960 |
root |
1.50 |
# elif __FreeBSD__ |
961 |
root |
1.69 |
/* |
962 |
|
|
* Of course, the freebsd sendfile is a dire hack with no thoughts |
963 |
|
|
* wasted on making it similar to other I/O functions. |
964 |
|
|
*/ |
965 |
|
|
off_t sbytes; |
966 |
|
|
res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
967 |
|
|
|
968 |
|
|
#if 0 /* according to the manpage, this is correct, but broken behaviour */ |
969 |
|
|
/* freebsd' sendfile will return 0 on success */ |
970 |
|
|
/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
971 |
|
|
/* not on e.g. EIO or EPIPE - sounds broken */ |
972 |
|
|
if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
973 |
|
|
res = sbytes; |
974 |
|
|
#endif |
975 |
|
|
|
976 |
|
|
/* according to source inspection, this is correct, and useful behaviour */ |
977 |
|
|
if (sbytes) |
978 |
|
|
res = sbytes; |
979 |
root |
1.1 |
|
980 |
root |
1.46 |
# elif defined (__APPLE__) |
981 |
root |
1.69 |
off_t sbytes = count; |
982 |
|
|
res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
983 |
root |
1.45 |
|
984 |
root |
1.69 |
/* according to the manpage, sbytes is always valid */ |
985 |
|
|
if (sbytes) |
986 |
|
|
res = sbytes; |
987 |
root |
1.45 |
|
988 |
root |
1.1 |
# elif __hpux |
989 |
root |
1.69 |
res = sendfile (ofd, ifd, offset, count, 0, 0); |
990 |
root |
1.1 |
|
991 |
|
|
# elif __solaris |
992 |
root |
1.69 |
struct sendfilevec vec; |
993 |
|
|
size_t sbytes; |
994 |
root |
1.1 |
|
995 |
root |
1.69 |
vec.sfv_fd = ifd; |
996 |
|
|
vec.sfv_flag = 0; |
997 |
|
|
vec.sfv_off = offset; |
998 |
|
|
vec.sfv_len = count; |
999 |
root |
1.1 |
|
1000 |
root |
1.69 |
res = sendfilev (ofd, &vec, 1, &sbytes); |
1001 |
root |
1.1 |
|
1002 |
root |
1.69 |
if (res < 0 && sbytes) |
1003 |
|
|
res = sbytes; |
1004 |
root |
1.1 |
|
1005 |
|
|
# endif |
1006 |
root |
1.46 |
|
1007 |
|
|
#elif defined (_WIN32) |
1008 |
root |
1.69 |
/* does not work, just for documentation of what would need to be done */ |
1009 |
root |
1.70 |
/* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
1010 |
|
|
/* libeio guarantees that the file offset does not change, and windows */ |
1011 |
|
|
/* has no way to get an independent handle to the same file description */ |
1012 |
root |
1.69 |
HANDLE h = TO_SOCKET (ifd); |
1013 |
|
|
SetFilePointer (h, offset, 0, FILE_BEGIN); |
1014 |
|
|
res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
1015 |
root |
1.45 |
|
1016 |
root |
1.1 |
#else |
1017 |
root |
1.69 |
res = -1; |
1018 |
|
|
errno = ENOSYS; |
1019 |
root |
1.1 |
#endif |
1020 |
|
|
|
1021 |
root |
1.69 |
/* we assume sendfile can copy at least 128mb in one go */ |
1022 |
|
|
if (res <= 128 * 1024 * 1024) |
1023 |
|
|
{ |
1024 |
|
|
if (res > 0) |
1025 |
|
|
written += res; |
1026 |
|
|
|
1027 |
|
|
if (written) |
1028 |
|
|
return written; |
1029 |
|
|
|
1030 |
|
|
break; |
1031 |
|
|
} |
1032 |
|
|
else |
1033 |
|
|
{ |
1034 |
|
|
/* if we requested more, then probably the kernel was lazy */ |
1035 |
|
|
written += res; |
1036 |
|
|
offset += res; |
1037 |
|
|
count -= res; |
1038 |
|
|
|
1039 |
|
|
if (!count) |
1040 |
|
|
return written; |
1041 |
|
|
} |
1042 |
|
|
} |
1043 |
|
|
|
1044 |
|
|
if (res < 0 |
1045 |
root |
1.1 |
&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1046 |
root |
1.49 |
/* BSDs */ |
1047 |
|
|
#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1048 |
|
|
|| errno == ENOTSUP |
1049 |
|
|
#endif |
1050 |
|
|
|| errno == EOPNOTSUPP /* BSDs */ |
1051 |
root |
1.1 |
#if __solaris |
1052 |
|
|
|| errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1053 |
|
|
#endif |
1054 |
|
|
) |
1055 |
|
|
) |
1056 |
|
|
{ |
1057 |
|
|
/* emulate sendfile. this is a major pain in the ass */ |
1058 |
|
|
dBUF; |
1059 |
|
|
|
1060 |
|
|
res = 0; |
1061 |
|
|
|
1062 |
|
|
while (count) |
1063 |
|
|
{ |
1064 |
|
|
ssize_t cnt; |
1065 |
|
|
|
1066 |
|
|
cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1067 |
|
|
|
1068 |
|
|
if (cnt <= 0) |
1069 |
|
|
{ |
1070 |
|
|
if (cnt && !res) res = -1; |
1071 |
|
|
break; |
1072 |
|
|
} |
1073 |
|
|
|
1074 |
|
|
cnt = write (ofd, eio_buf, cnt); |
1075 |
|
|
|
1076 |
|
|
if (cnt <= 0) |
1077 |
|
|
{ |
1078 |
|
|
if (cnt && !res) res = -1; |
1079 |
|
|
break; |
1080 |
|
|
} |
1081 |
|
|
|
1082 |
|
|
offset += cnt; |
1083 |
|
|
res += cnt; |
1084 |
|
|
count -= cnt; |
1085 |
|
|
} |
1086 |
|
|
} |
1087 |
|
|
|
1088 |
|
|
return res; |
1089 |
|
|
} |
1090 |
|
|
|
1091 |
root |
1.37 |
static signed char |
1092 |
|
|
eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1093 |
|
|
{ |
1094 |
root |
1.63 |
return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1095 |
|
|
: a->inode < b->inode ? -1 |
1096 |
|
|
: a->inode > b->inode ? 1 |
1097 |
|
|
: 0; |
1098 |
root |
1.37 |
} |
1099 |
|
|
|
1100 |
root |
1.39 |
#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1101 |
|
|
|
1102 |
root |
1.42 |
#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1103 |
|
|
#define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1104 |
root |
1.37 |
|
1105 |
|
|
static void |
1106 |
root |
1.42 |
eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1107 |
root |
1.31 |
{ |
1108 |
root |
1.42 |
unsigned char bits [9 + sizeof (ino_t) * 8]; |
1109 |
|
|
unsigned char *bit = bits; |
1110 |
|
|
|
1111 |
|
|
assert (CHAR_BIT == 8); |
1112 |
|
|
assert (sizeof (eio_dirent) * 8 < 256); |
1113 |
root |
1.63 |
assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1114 |
|
|
assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
1115 |
root |
1.42 |
|
1116 |
|
|
if (size <= EIO_SORT_FAST) |
1117 |
|
|
return; |
1118 |
|
|
|
1119 |
|
|
/* first prepare an array of bits to test in our radix sort */ |
1120 |
|
|
/* try to take endianness into account, as well as differences in ino_t sizes */ |
1121 |
|
|
/* inode_bits must contain all inodes ORed together */ |
1122 |
|
|
/* which is used to skip bits that are 0 everywhere, which is very common */ |
1123 |
|
|
{ |
1124 |
|
|
ino_t endianness; |
1125 |
|
|
int i, j; |
1126 |
root |
1.39 |
|
1127 |
root |
1.42 |
/* we store the byte offset of byte n into byte n of "endianness" */ |
1128 |
|
|
for (i = 0; i < sizeof (ino_t); ++i) |
1129 |
|
|
((unsigned char *)&endianness)[i] = i; |
1130 |
root |
1.37 |
|
1131 |
root |
1.42 |
*bit++ = 0; |
1132 |
root |
1.37 |
|
1133 |
root |
1.42 |
for (i = 0; i < sizeof (ino_t); ++i) |
1134 |
|
|
{ |
1135 |
|
|
/* shifting off the byte offsets out of "endianness" */ |
1136 |
|
|
int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1137 |
|
|
endianness >>= 8; |
1138 |
|
|
|
1139 |
|
|
for (j = 0; j < 8; ++j) |
1140 |
|
|
if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
1141 |
|
|
*bit++ = offs + j; |
1142 |
|
|
} |
1143 |
root |
1.38 |
|
1144 |
root |
1.42 |
for (j = 0; j < 8; ++j) |
1145 |
|
|
if (score_bits & (1 << j)) |
1146 |
|
|
*bit++ = offsetof (eio_dirent, score) * 8 + j; |
1147 |
|
|
} |
1148 |
root |
1.37 |
|
1149 |
root |
1.42 |
/* now actually do the sorting (a variant of MSD radix sort) */ |
1150 |
|
|
{ |
1151 |
|
|
eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
1152 |
|
|
eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
1153 |
|
|
unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
1154 |
|
|
int stk_idx = 0; |
1155 |
|
|
|
1156 |
|
|
base_stk [stk_idx] = dents; |
1157 |
|
|
end_stk [stk_idx] = dents + size; |
1158 |
|
|
bit_stk [stk_idx] = bit - 1; |
1159 |
root |
1.37 |
|
1160 |
root |
1.42 |
do |
1161 |
|
|
{ |
1162 |
|
|
base = base_stk [stk_idx]; |
1163 |
|
|
end = end_stk [stk_idx]; |
1164 |
|
|
bit = bit_stk [stk_idx]; |
1165 |
root |
1.37 |
|
1166 |
root |
1.42 |
for (;;) |
1167 |
|
|
{ |
1168 |
|
|
unsigned char O = *bit >> 3; |
1169 |
|
|
unsigned char M = 1 << (*bit & 7); |
1170 |
root |
1.31 |
|
1171 |
root |
1.42 |
eio_dirent *a = base; |
1172 |
|
|
eio_dirent *b = end; |
1173 |
root |
1.37 |
|
1174 |
root |
1.42 |
if (b - a < EIO_SORT_CUTOFF) |
1175 |
|
|
break; |
1176 |
root |
1.37 |
|
1177 |
root |
1.42 |
/* now bit-partition the array on the bit */ |
1178 |
|
|
/* this ugly asymmetric loop seems to perform much better than typical */ |
1179 |
|
|
/* partition algos found in the literature */ |
1180 |
|
|
do |
1181 |
|
|
if (!(((unsigned char *)a)[O] & M)) |
1182 |
|
|
++a; |
1183 |
|
|
else if (!(((unsigned char *)--b)[O] & M)) |
1184 |
root |
1.38 |
{ |
1185 |
root |
1.42 |
eio_dirent tmp = *a; *a = *b; *b = tmp; |
1186 |
|
|
++a; |
1187 |
|
|
} |
1188 |
|
|
while (b > a); |
1189 |
|
|
|
1190 |
|
|
/* next bit, or stop, if no bits left in this path */ |
1191 |
|
|
if (!*--bit) |
1192 |
|
|
break; |
1193 |
root |
1.37 |
|
1194 |
root |
1.42 |
base_stk [stk_idx] = a; |
1195 |
|
|
end_stk [stk_idx] = end; |
1196 |
|
|
bit_stk [stk_idx] = bit; |
1197 |
|
|
++stk_idx; |
1198 |
|
|
|
1199 |
|
|
end = a; |
1200 |
|
|
} |
1201 |
|
|
} |
1202 |
|
|
while (stk_idx--); |
1203 |
|
|
} |
1204 |
|
|
} |
1205 |
root |
1.37 |
|
1206 |
root |
1.42 |
static void |
1207 |
|
|
eio_dent_insertion_sort (eio_dirent *dents, int size) |
1208 |
|
|
{ |
1209 |
root |
1.39 |
/* first move the smallest element to the front, to act as a sentinel */ |
1210 |
|
|
{ |
1211 |
root |
1.41 |
int i; |
1212 |
|
|
eio_dirent *min = dents; |
1213 |
|
|
|
1214 |
root |
1.42 |
/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
1215 |
|
|
for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
1216 |
root |
1.41 |
if (EIO_DENT_CMP (dents [i], <, *min)) |
1217 |
|
|
min = &dents [i]; |
1218 |
root |
1.39 |
|
1219 |
|
|
/* swap elements 0 and j (minimum) */ |
1220 |
|
|
{ |
1221 |
root |
1.41 |
eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
1222 |
root |
1.39 |
} |
1223 |
|
|
} |
1224 |
|
|
|
1225 |
root |
1.42 |
/* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
1226 |
root |
1.41 |
{ |
1227 |
|
|
eio_dirent *i, *j; |
1228 |
|
|
|
1229 |
|
|
for (i = dents + 1; i < dents + size; ++i) |
1230 |
|
|
{ |
1231 |
|
|
eio_dirent value = *i; |
1232 |
root |
1.37 |
|
1233 |
root |
1.41 |
for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
1234 |
|
|
j [1] = j [0]; |
1235 |
root |
1.37 |
|
1236 |
root |
1.41 |
j [1] = value; |
1237 |
|
|
} |
1238 |
|
|
} |
1239 |
root |
1.31 |
} |
1240 |
|
|
|
1241 |
root |
1.42 |
static void |
1242 |
|
|
eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1243 |
|
|
{ |
1244 |
|
|
if (size <= 1) |
1245 |
|
|
return; /* our insertion sort relies on size > 0 */ |
1246 |
|
|
|
1247 |
|
|
/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1248 |
|
|
/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
1249 |
|
|
eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
1250 |
|
|
|
1251 |
|
|
/* use an insertion sort at the end, or for small arrays, */ |
1252 |
|
|
/* as insertion sort is more efficient for small partitions */ |
1253 |
|
|
eio_dent_insertion_sort (dents, size); |
1254 |
|
|
} |
1255 |
|
|
|
1256 |
root |
1.1 |
/* read a full directory */ |
1257 |
root |
1.9 |
static void |
1258 |
root |
1.14 |
eio__scandir (eio_req *req, etp_worker *self) |
1259 |
root |
1.1 |
{ |
1260 |
|
|
DIR *dirp; |
1261 |
|
|
EIO_STRUCT_DIRENT *entp; |
1262 |
root |
1.37 |
char *name, *names; |
1263 |
root |
1.31 |
int namesalloc = 4096; |
1264 |
|
|
int namesoffs = 0; |
1265 |
|
|
int flags = req->int1; |
1266 |
|
|
eio_dirent *dents = 0; |
1267 |
|
|
int dentalloc = 128; |
1268 |
|
|
int dentoffs = 0; |
1269 |
root |
1.42 |
ino_t inode_bits = 0; |
1270 |
root |
1.31 |
|
1271 |
|
|
req->result = -1; |
1272 |
|
|
|
1273 |
|
|
if (!(flags & EIO_READDIR_DENTS)) |
1274 |
|
|
flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1275 |
root |
1.1 |
|
1276 |
root |
1.14 |
X_LOCK (wrklock); |
1277 |
root |
1.20 |
/* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1278 |
root |
1.1 |
self->dirp = dirp = opendir (req->ptr1); |
1279 |
root |
1.63 |
|
1280 |
root |
1.31 |
req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1281 |
root |
1.42 |
req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1282 |
|
|
req->ptr2 = names = malloc (namesalloc); |
1283 |
root |
1.14 |
X_UNLOCK (wrklock); |
1284 |
root |
1.1 |
|
1285 |
root |
1.31 |
if (dirp && names && (!flags || dents)) |
1286 |
root |
1.1 |
for (;;) |
1287 |
|
|
{ |
1288 |
|
|
errno = 0; |
1289 |
root |
1.16 |
entp = readdir (dirp); |
1290 |
root |
1.1 |
|
1291 |
|
|
if (!entp) |
1292 |
root |
1.31 |
{ |
1293 |
|
|
if (errno) |
1294 |
|
|
break; |
1295 |
|
|
|
1296 |
|
|
/* sort etc. */ |
1297 |
|
|
req->int1 = flags; |
1298 |
|
|
req->result = dentoffs; |
1299 |
|
|
|
1300 |
root |
1.42 |
if (flags & EIO_READDIR_STAT_ORDER) |
1301 |
root |
1.63 |
eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1302 |
root |
1.32 |
else if (flags & EIO_READDIR_DIRS_FIRST) |
1303 |
root |
1.42 |
if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1304 |
|
|
eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1305 |
|
|
else |
1306 |
|
|
{ |
1307 |
|
|
/* in this case, all is known, and we just put dirs first and sort them */ |
1308 |
|
|
eio_dirent *oth = dents + dentoffs; |
1309 |
|
|
eio_dirent *dir = dents; |
1310 |
|
|
|
1311 |
|
|
/* now partition dirs to the front, and non-dirs to the back */ |
1312 |
|
|
/* by walking from both sides and swapping if necessary */ |
1313 |
|
|
while (oth > dir) |
1314 |
|
|
{ |
1315 |
|
|
if (dir->type == EIO_DT_DIR) |
1316 |
root |
1.41 |
++dir; |
1317 |
root |
1.42 |
else if ((--oth)->type == EIO_DT_DIR) |
1318 |
|
|
{ |
1319 |
|
|
eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1320 |
|
|
|
1321 |
|
|
++dir; |
1322 |
|
|
} |
1323 |
|
|
} |
1324 |
root |
1.31 |
|
1325 |
root |
1.64 |
/* now sort the dirs only (dirs all have the same score) */ |
1326 |
root |
1.42 |
eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1327 |
|
|
} |
1328 |
root |
1.31 |
|
1329 |
|
|
break; |
1330 |
|
|
} |
1331 |
root |
1.1 |
|
1332 |
root |
1.32 |
/* now add the entry to our list(s) */ |
1333 |
root |
1.1 |
name = entp->d_name; |
1334 |
|
|
|
1335 |
root |
1.32 |
/* skip . and .. entries */ |
1336 |
root |
1.1 |
if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1337 |
|
|
{ |
1338 |
root |
1.33 |
int len = D_NAMLEN (entp) + 1; |
1339 |
root |
1.1 |
|
1340 |
root |
1.31 |
while (expect_false (namesoffs + len > namesalloc)) |
1341 |
root |
1.1 |
{ |
1342 |
root |
1.31 |
namesalloc *= 2; |
1343 |
root |
1.14 |
X_LOCK (wrklock); |
1344 |
root |
1.42 |
req->ptr2 = names = realloc (names, namesalloc); |
1345 |
root |
1.14 |
X_UNLOCK (wrklock); |
1346 |
root |
1.1 |
|
1347 |
|
|
if (!names) |
1348 |
|
|
break; |
1349 |
|
|
} |
1350 |
|
|
|
1351 |
root |
1.31 |
memcpy (names + namesoffs, name, len); |
1352 |
|
|
|
1353 |
|
|
if (dents) |
1354 |
|
|
{ |
1355 |
|
|
struct eio_dirent *ent; |
1356 |
|
|
|
1357 |
|
|
if (expect_false (dentoffs == dentalloc)) |
1358 |
|
|
{ |
1359 |
|
|
dentalloc *= 2; |
1360 |
|
|
X_LOCK (wrklock); |
1361 |
root |
1.42 |
req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1362 |
root |
1.31 |
X_UNLOCK (wrklock); |
1363 |
|
|
|
1364 |
|
|
if (!dents) |
1365 |
|
|
break; |
1366 |
|
|
} |
1367 |
|
|
|
1368 |
|
|
ent = dents + dentoffs; |
1369 |
|
|
|
1370 |
root |
1.42 |
ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1371 |
root |
1.31 |
ent->namelen = len - 1; |
1372 |
root |
1.42 |
ent->inode = D_INO (entp); |
1373 |
|
|
|
1374 |
|
|
inode_bits |= ent->inode; |
1375 |
root |
1.31 |
|
1376 |
|
|
switch (D_TYPE (entp)) |
1377 |
|
|
{ |
1378 |
|
|
default: |
1379 |
|
|
ent->type = EIO_DT_UNKNOWN; |
1380 |
|
|
flags |= EIO_READDIR_FOUND_UNKNOWN; |
1381 |
|
|
break; |
1382 |
|
|
|
1383 |
|
|
#ifdef DT_FIFO |
1384 |
|
|
case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1385 |
|
|
#endif |
1386 |
|
|
#ifdef DT_CHR |
1387 |
|
|
case DT_CHR: ent->type = EIO_DT_CHR; break; |
1388 |
|
|
#endif |
1389 |
root |
1.33 |
#ifdef DT_MPC |
1390 |
|
|
case DT_MPC: ent->type = EIO_DT_MPC; break; |
1391 |
|
|
#endif |
1392 |
root |
1.32 |
#ifdef DT_DIR |
1393 |
root |
1.31 |
case DT_DIR: ent->type = EIO_DT_DIR; break; |
1394 |
|
|
#endif |
1395 |
root |
1.33 |
#ifdef DT_NAM |
1396 |
|
|
case DT_NAM: ent->type = EIO_DT_NAM; break; |
1397 |
|
|
#endif |
1398 |
root |
1.32 |
#ifdef DT_BLK |
1399 |
root |
1.31 |
case DT_BLK: ent->type = EIO_DT_BLK; break; |
1400 |
|
|
#endif |
1401 |
root |
1.33 |
#ifdef DT_MPB |
1402 |
|
|
case DT_MPB: ent->type = EIO_DT_MPB; break; |
1403 |
|
|
#endif |
1404 |
root |
1.32 |
#ifdef DT_REG |
1405 |
root |
1.31 |
case DT_REG: ent->type = EIO_DT_REG; break; |
1406 |
|
|
#endif |
1407 |
root |
1.33 |
#ifdef DT_NWK |
1408 |
|
|
case DT_NWK: ent->type = EIO_DT_NWK; break; |
1409 |
|
|
#endif |
1410 |
|
|
#ifdef DT_CMP |
1411 |
|
|
case DT_CMP: ent->type = EIO_DT_CMP; break; |
1412 |
|
|
#endif |
1413 |
root |
1.32 |
#ifdef DT_LNK |
1414 |
root |
1.31 |
case DT_LNK: ent->type = EIO_DT_LNK; break; |
1415 |
|
|
#endif |
1416 |
|
|
#ifdef DT_SOCK |
1417 |
|
|
case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1418 |
|
|
#endif |
1419 |
root |
1.33 |
#ifdef DT_DOOR |
1420 |
|
|
case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
1421 |
|
|
#endif |
1422 |
root |
1.31 |
#ifdef DT_WHT |
1423 |
|
|
case DT_WHT: ent->type = EIO_DT_WHT; break; |
1424 |
|
|
#endif |
1425 |
|
|
} |
1426 |
|
|
|
1427 |
root |
1.42 |
ent->score = 7; |
1428 |
root |
1.31 |
|
1429 |
|
|
if (flags & EIO_READDIR_DIRS_FIRST) |
1430 |
|
|
{ |
1431 |
|
|
if (ent->type == EIO_DT_UNKNOWN) |
1432 |
|
|
{ |
1433 |
|
|
if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1434 |
root |
1.42 |
ent->score = 1; |
1435 |
root |
1.31 |
else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1436 |
root |
1.42 |
ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1437 |
root |
1.31 |
} |
1438 |
root |
1.33 |
else if (ent->type == EIO_DT_DIR) |
1439 |
root |
1.42 |
ent->score = 0; |
1440 |
root |
1.31 |
} |
1441 |
|
|
} |
1442 |
|
|
|
1443 |
|
|
namesoffs += len; |
1444 |
|
|
++dentoffs; |
1445 |
root |
1.1 |
} |
1446 |
root |
1.36 |
|
1447 |
|
|
if (EIO_CANCELLED (req)) |
1448 |
|
|
{ |
1449 |
|
|
errno = ECANCELED; |
1450 |
|
|
break; |
1451 |
|
|
} |
1452 |
root |
1.1 |
} |
1453 |
|
|
} |
1454 |
|
|
|
1455 |
root |
1.56 |
#ifdef PAGESIZE |
1456 |
|
|
# define eio_pagesize() PAGESIZE |
1457 |
|
|
#else |
1458 |
|
|
static intptr_t |
1459 |
|
|
eio_pagesize (void) |
1460 |
|
|
{ |
1461 |
|
|
static intptr_t page; |
1462 |
|
|
|
1463 |
|
|
if (!page) |
1464 |
|
|
page = sysconf (_SC_PAGESIZE); |
1465 |
|
|
|
1466 |
|
|
return page; |
1467 |
|
|
} |
1468 |
|
|
#endif |
1469 |
|
|
|
1470 |
|
|
static void |
1471 |
|
|
eio_page_align (void **addr, size_t *length) |
1472 |
|
|
{ |
1473 |
|
|
intptr_t mask = eio_pagesize () - 1; |
1474 |
|
|
|
1475 |
|
|
/* round down addr */ |
1476 |
|
|
intptr_t adj = mask & (intptr_t)*addr; |
1477 |
|
|
|
1478 |
|
|
*addr = (void *)((intptr_t)*addr - adj); |
1479 |
|
|
*length += adj; |
1480 |
|
|
|
1481 |
|
|
/* round up length */ |
1482 |
|
|
*length = (*length + mask) & ~mask; |
1483 |
|
|
} |
1484 |
|
|
|
1485 |
|
|
#if !_POSIX_MEMLOCK |
1486 |
|
|
# define eio__mlockall(a) ((errno = ENOSYS), -1) |
1487 |
|
|
#else |
1488 |
|
|
|
1489 |
|
|
static int |
1490 |
|
|
eio__mlockall (int flags) |
1491 |
|
|
{ |
1492 |
|
|
#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
1493 |
|
|
extern int mallopt (int, int); |
1494 |
|
|
mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
1495 |
|
|
#endif |
1496 |
|
|
|
1497 |
|
|
if (EIO_MCL_CURRENT != MCL_CURRENT |
1498 |
|
|
|| EIO_MCL_FUTURE != MCL_FUTURE) |
1499 |
|
|
{ |
1500 |
|
|
flags = 0 |
1501 |
|
|
| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
1502 |
|
|
| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
1503 |
|
|
} |
1504 |
|
|
|
1505 |
root |
1.57 |
return mlockall (flags); |
1506 |
root |
1.56 |
} |
1507 |
|
|
#endif |
1508 |
|
|
|
1509 |
root |
1.60 |
#if !_POSIX_MEMLOCK_RANGE |
1510 |
|
|
# define eio__mlock(a,b) ((errno = ENOSYS), -1) |
1511 |
|
|
#else |
1512 |
|
|
|
1513 |
|
|
static int |
1514 |
|
|
eio__mlock (void *addr, size_t length) |
1515 |
|
|
{ |
1516 |
|
|
eio_page_align (&addr, &length); |
1517 |
|
|
|
1518 |
|
|
return mlock (addr, length); |
1519 |
|
|
} |
1520 |
|
|
|
1521 |
|
|
#endif |
1522 |
|
|
|
1523 |
root |
1.23 |
#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1524 |
root |
1.48 |
# define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1525 |
|
|
#else |
1526 |
|
|
|
1527 |
|
|
int |
1528 |
|
|
eio__msync (void *mem, size_t len, int flags) |
1529 |
|
|
{ |
1530 |
root |
1.56 |
eio_page_align (&mem, &len); |
1531 |
|
|
|
1532 |
root |
1.48 |
if (EIO_MS_ASYNC != MS_SYNC |
1533 |
|
|
|| EIO_MS_INVALIDATE != MS_INVALIDATE |
1534 |
|
|
|| EIO_MS_SYNC != MS_SYNC) |
1535 |
|
|
{ |
1536 |
|
|
flags = 0 |
1537 |
|
|
| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
1538 |
|
|
| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
1539 |
|
|
| (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
1540 |
|
|
} |
1541 |
|
|
|
1542 |
|
|
return msync (mem, len, flags); |
1543 |
|
|
} |
1544 |
|
|
|
1545 |
root |
1.23 |
#endif |
1546 |
|
|
|
1547 |
|
|
int |
1548 |
root |
1.67 |
eio__mtouch (eio_req *req) |
1549 |
root |
1.23 |
{ |
1550 |
root |
1.67 |
void *mem = req->ptr2; |
1551 |
|
|
size_t len = req->size; |
1552 |
|
|
int flags = req->int1; |
1553 |
|
|
|
1554 |
root |
1.56 |
eio_page_align (&mem, &len); |
1555 |
root |
1.23 |
|
1556 |
root |
1.56 |
{ |
1557 |
|
|
intptr_t addr = (intptr_t)mem; |
1558 |
|
|
intptr_t end = addr + len; |
1559 |
|
|
intptr_t page = eio_pagesize (); |
1560 |
|
|
|
1561 |
|
|
if (addr < end) |
1562 |
|
|
if (flags & EIO_MT_MODIFY) /* modify */ |
1563 |
root |
1.67 |
do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
1564 |
root |
1.56 |
else |
1565 |
root |
1.67 |
do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
1566 |
root |
1.56 |
} |
1567 |
root |
1.23 |
|
1568 |
|
|
return 0; |
1569 |
|
|
} |
1570 |
|
|
|
1571 |
root |
1.1 |
/*****************************************************************************/ |
1572 |
|
|
|
1573 |
root |
1.6 |
#define ALLOC(len) \ |
1574 |
|
|
if (!req->ptr2) \ |
1575 |
|
|
{ \ |
1576 |
root |
1.14 |
X_LOCK (wrklock); \ |
1577 |
root |
1.7 |
req->flags |= EIO_FLAG_PTR2_FREE; \ |
1578 |
root |
1.14 |
X_UNLOCK (wrklock); \ |
1579 |
root |
1.7 |
req->ptr2 = malloc (len); \ |
1580 |
|
|
if (!req->ptr2) \ |
1581 |
|
|
{ \ |
1582 |
|
|
errno = ENOMEM; \ |
1583 |
|
|
req->result = -1; \ |
1584 |
|
|
break; \ |
1585 |
|
|
} \ |
1586 |
root |
1.6 |
} |
1587 |
|
|
|
1588 |
root |
1.14 |
X_THREAD_PROC (etp_proc) |
1589 |
root |
1.1 |
{ |
1590 |
root |
1.14 |
ETP_REQ *req; |
1591 |
root |
1.1 |
struct timespec ts; |
1592 |
root |
1.14 |
etp_worker *self = (etp_worker *)thr_arg; |
1593 |
root |
1.1 |
|
1594 |
|
|
/* try to distribute timeouts somewhat randomly */ |
1595 |
|
|
ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1596 |
|
|
|
1597 |
|
|
for (;;) |
1598 |
|
|
{ |
1599 |
|
|
X_LOCK (reqlock); |
1600 |
|
|
|
1601 |
|
|
for (;;) |
1602 |
|
|
{ |
1603 |
|
|
self->req = req = reqq_shift (&req_queue); |
1604 |
|
|
|
1605 |
|
|
if (req) |
1606 |
|
|
break; |
1607 |
|
|
|
1608 |
|
|
++idle; |
1609 |
|
|
|
1610 |
root |
1.62 |
ts.tv_sec = time (0) + idle_timeout; |
1611 |
root |
1.1 |
if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1612 |
|
|
{ |
1613 |
|
|
if (idle > max_idle) |
1614 |
|
|
{ |
1615 |
|
|
--idle; |
1616 |
|
|
X_UNLOCK (reqlock); |
1617 |
root |
1.14 |
X_LOCK (wrklock); |
1618 |
root |
1.1 |
--started; |
1619 |
root |
1.14 |
X_UNLOCK (wrklock); |
1620 |
root |
1.1 |
goto quit; |
1621 |
|
|
} |
1622 |
|
|
|
1623 |
|
|
/* we are allowed to idle, so do so without any timeout */ |
1624 |
|
|
X_COND_WAIT (reqwait, reqlock); |
1625 |
|
|
} |
1626 |
|
|
|
1627 |
|
|
--idle; |
1628 |
|
|
} |
1629 |
|
|
|
1630 |
|
|
--nready; |
1631 |
|
|
|
1632 |
|
|
X_UNLOCK (reqlock); |
1633 |
|
|
|
1634 |
root |
1.15 |
if (req->type < 0) |
1635 |
|
|
goto quit; |
1636 |
root |
1.1 |
|
1637 |
|
|
if (!EIO_CANCELLED (req)) |
1638 |
root |
1.15 |
ETP_EXECUTE (self, req); |
1639 |
root |
1.1 |
|
1640 |
|
|
X_LOCK (reslock); |
1641 |
|
|
|
1642 |
|
|
++npending; |
1643 |
|
|
|
1644 |
root |
1.14 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
1645 |
|
|
want_poll_cb (); |
1646 |
root |
1.1 |
|
1647 |
|
|
self->req = 0; |
1648 |
root |
1.14 |
etp_worker_clear (self); |
1649 |
root |
1.1 |
|
1650 |
|
|
X_UNLOCK (reslock); |
1651 |
|
|
} |
1652 |
|
|
|
1653 |
|
|
quit: |
1654 |
root |
1.14 |
X_LOCK (wrklock); |
1655 |
|
|
etp_worker_free (self); |
1656 |
|
|
X_UNLOCK (wrklock); |
1657 |
root |
1.1 |
|
1658 |
|
|
return 0; |
1659 |
|
|
} |
1660 |
|
|
|
1661 |
|
|
/*****************************************************************************/ |
1662 |
|
|
|
1663 |
|
|
int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1664 |
|
|
{ |
1665 |
root |
1.19 |
return etp_init (want_poll, done_poll); |
1666 |
root |
1.1 |
} |
1667 |
|
|
|
1668 |
root |
1.6 |
static void eio_api_destroy (eio_req *req) |
1669 |
|
|
{ |
1670 |
|
|
free (req); |
1671 |
|
|
} |
1672 |
|
|
|
1673 |
|
|
#define REQ(rtype) \ |
1674 |
|
|
eio_req *req; \ |
1675 |
|
|
\ |
1676 |
|
|
req = (eio_req *)calloc (1, sizeof *req); \ |
1677 |
|
|
if (!req) \ |
1678 |
|
|
return 0; \ |
1679 |
|
|
\ |
1680 |
root |
1.11 |
req->type = rtype; \ |
1681 |
|
|
req->pri = pri; \ |
1682 |
|
|
req->finish = cb; \ |
1683 |
|
|
req->data = data; \ |
1684 |
root |
1.6 |
req->destroy = eio_api_destroy; |
1685 |
|
|
|
1686 |
|
|
#define SEND eio_submit (req); return req |
1687 |
|
|
|
1688 |
root |
1.9 |
#define PATH \ |
1689 |
|
|
req->flags |= EIO_FLAG_PTR1_FREE; \ |
1690 |
|
|
req->ptr1 = strdup (path); \ |
1691 |
|
|
if (!req->ptr1) \ |
1692 |
|
|
{ \ |
1693 |
|
|
eio_api_destroy (req); \ |
1694 |
|
|
return 0; \ |
1695 |
|
|
} |
1696 |
|
|
|
1697 |
root |
1.15 |
static void eio_execute (etp_worker *self, eio_req *req) |
1698 |
|
|
{ |
1699 |
|
|
switch (req->type) |
1700 |
|
|
{ |
1701 |
|
|
case EIO_READ: ALLOC (req->size); |
1702 |
|
|
req->result = req->offs >= 0 |
1703 |
|
|
? pread (req->int1, req->ptr2, req->size, req->offs) |
1704 |
|
|
: read (req->int1, req->ptr2, req->size); break; |
1705 |
|
|
case EIO_WRITE: req->result = req->offs >= 0 |
1706 |
|
|
? pwrite (req->int1, req->ptr2, req->size, req->offs) |
1707 |
|
|
: write (req->int1, req->ptr2, req->size); break; |
1708 |
|
|
|
1709 |
|
|
case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
1710 |
|
|
case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
1711 |
|
|
|
1712 |
|
|
case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1713 |
|
|
req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1714 |
|
|
case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1715 |
|
|
req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1716 |
|
|
case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1717 |
|
|
req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1718 |
|
|
|
1719 |
root |
1.51 |
case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
1720 |
|
|
req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
1721 |
|
|
case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
1722 |
|
|
req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
1723 |
|
|
|
1724 |
root |
1.15 |
case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1725 |
|
|
case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1726 |
|
|
case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1727 |
|
|
case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1728 |
|
|
case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1729 |
|
|
case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
1730 |
|
|
|
1731 |
|
|
case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
1732 |
|
|
case EIO_CLOSE: req->result = close (req->int1); break; |
1733 |
|
|
case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
1734 |
|
|
case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
1735 |
|
|
case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1736 |
|
|
case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1737 |
|
|
case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1738 |
|
|
case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1739 |
|
|
case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1740 |
root |
1.61 |
case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1741 |
root |
1.15 |
|
1742 |
root |
1.54 |
case EIO_READLINK: ALLOC (PATH_MAX); |
1743 |
|
|
req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1744 |
root |
1.15 |
|
1745 |
|
|
case EIO_SYNC: req->result = 0; sync (); break; |
1746 |
|
|
case EIO_FSYNC: req->result = fsync (req->int1); break; |
1747 |
|
|
case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1748 |
root |
1.48 |
case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1749 |
root |
1.67 |
case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1750 |
root |
1.56 |
case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
1751 |
|
|
case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1752 |
root |
1.27 |
case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1753 |
root |
1.15 |
|
1754 |
|
|
case EIO_READDIR: eio__scandir (req, self); break; |
1755 |
|
|
|
1756 |
|
|
case EIO_BUSY: |
1757 |
|
|
#ifdef _WIN32 |
1758 |
root |
1.57 |
Sleep (req->nv1 * 1e3); |
1759 |
root |
1.15 |
#else |
1760 |
|
|
{ |
1761 |
|
|
struct timeval tv; |
1762 |
|
|
|
1763 |
|
|
tv.tv_sec = req->nv1; |
1764 |
root |
1.57 |
tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1765 |
root |
1.15 |
|
1766 |
|
|
req->result = select (0, 0, 0, 0, &tv); |
1767 |
|
|
} |
1768 |
|
|
#endif |
1769 |
|
|
break; |
1770 |
|
|
|
1771 |
|
|
case EIO_UTIME: |
1772 |
|
|
case EIO_FUTIME: |
1773 |
|
|
{ |
1774 |
|
|
struct timeval tv[2]; |
1775 |
|
|
struct timeval *times; |
1776 |
|
|
|
1777 |
|
|
if (req->nv1 != -1. || req->nv2 != -1.) |
1778 |
|
|
{ |
1779 |
|
|
tv[0].tv_sec = req->nv1; |
1780 |
|
|
tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
1781 |
|
|
tv[1].tv_sec = req->nv2; |
1782 |
|
|
tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
1783 |
|
|
|
1784 |
|
|
times = tv; |
1785 |
|
|
} |
1786 |
|
|
else |
1787 |
|
|
times = 0; |
1788 |
|
|
|
1789 |
|
|
req->result = req->type == EIO_FUTIME |
1790 |
|
|
? futimes (req->int1, times) |
1791 |
|
|
: utimes (req->ptr1, times); |
1792 |
|
|
} |
1793 |
root |
1.25 |
break; |
1794 |
root |
1.15 |
|
1795 |
|
|
case EIO_GROUP: |
1796 |
root |
1.24 |
abort (); /* handled in eio_request */ |
1797 |
|
|
|
1798 |
root |
1.15 |
case EIO_NOP: |
1799 |
|
|
req->result = 0; |
1800 |
|
|
break; |
1801 |
|
|
|
1802 |
|
|
case EIO_CUSTOM: |
1803 |
root |
1.68 |
req->feed (req); |
1804 |
root |
1.15 |
break; |
1805 |
|
|
|
1806 |
|
|
default: |
1807 |
root |
1.58 |
errno = ENOSYS; |
1808 |
root |
1.15 |
req->result = -1; |
1809 |
|
|
break; |
1810 |
|
|
} |
1811 |
|
|
|
1812 |
|
|
req->errorno = errno; |
1813 |
|
|
} |
1814 |
|
|
|
1815 |
root |
1.12 |
#ifndef EIO_NO_WRAPPERS |
1816 |
|
|
|
1817 |
root |
1.10 |
eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1818 |
root |
1.9 |
{ |
1819 |
|
|
REQ (EIO_NOP); SEND; |
1820 |
|
|
} |
1821 |
|
|
|
1822 |
root |
1.10 |
eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data) |
1823 |
root |
1.9 |
{ |
1824 |
|
|
REQ (EIO_BUSY); req->nv1 = delay; SEND; |
1825 |
|
|
} |
1826 |
|
|
|
1827 |
root |
1.10 |
eio_req *eio_sync (int pri, eio_cb cb, void *data) |
1828 |
root |
1.9 |
{ |
1829 |
|
|
REQ (EIO_SYNC); SEND; |
1830 |
|
|
} |
1831 |
root |
1.6 |
|
1832 |
root |
1.10 |
eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1833 |
root |
1.6 |
{ |
1834 |
root |
1.9 |
REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1835 |
root |
1.6 |
} |
1836 |
|
|
|
1837 |
root |
1.23 |
eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1838 |
|
|
{ |
1839 |
|
|
REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1840 |
|
|
} |
1841 |
|
|
|
1842 |
|
|
eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1843 |
|
|
{ |
1844 |
|
|
REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1845 |
|
|
} |
1846 |
|
|
|
1847 |
root |
1.56 |
eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
1848 |
|
|
{ |
1849 |
|
|
REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
1850 |
|
|
} |
1851 |
|
|
|
1852 |
|
|
eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
1853 |
|
|
{ |
1854 |
|
|
REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
1855 |
|
|
} |
1856 |
|
|
|
1857 |
root |
1.27 |
eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1858 |
|
|
{ |
1859 |
|
|
REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1860 |
|
|
} |
1861 |
|
|
|
1862 |
root |
1.10 |
eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1863 |
root |
1.6 |
{ |
1864 |
root |
1.9 |
REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1865 |
|
|
} |
1866 |
|
|
|
1867 |
root |
1.10 |
eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
1868 |
root |
1.9 |
{ |
1869 |
|
|
REQ (EIO_CLOSE); req->int1 = fd; SEND; |
1870 |
root |
1.6 |
} |
1871 |
|
|
|
1872 |
root |
1.10 |
eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) |
1873 |
root |
1.6 |
{ |
1874 |
root |
1.9 |
REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; |
1875 |
root |
1.6 |
} |
1876 |
|
|
|
1877 |
root |
1.10 |
eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
1878 |
root |
1.6 |
{ |
1879 |
root |
1.10 |
REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
1880 |
root |
1.6 |
} |
1881 |
|
|
|
1882 |
root |
1.10 |
eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
1883 |
root |
1.6 |
{ |
1884 |
root |
1.10 |
REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
1885 |
root |
1.6 |
} |
1886 |
|
|
|
1887 |
root |
1.10 |
eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1888 |
root |
1.6 |
{ |
1889 |
root |
1.9 |
REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1890 |
root |
1.6 |
} |
1891 |
|
|
|
1892 |
root |
1.51 |
eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
1893 |
|
|
{ |
1894 |
|
|
REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1895 |
|
|
} |
1896 |
|
|
|
1897 |
root |
1.10 |
eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1898 |
root |
1.6 |
{ |
1899 |
root |
1.9 |
REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1900 |
root |
1.6 |
} |
1901 |
|
|
|
1902 |
root |
1.10 |
eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data) |
1903 |
root |
1.6 |
{ |
1904 |
root |
1.9 |
REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; |
1905 |
root |
1.6 |
} |
1906 |
|
|
|
1907 |
root |
1.10 |
eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
1908 |
root |
1.6 |
{ |
1909 |
root |
1.9 |
REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
1910 |
root |
1.6 |
} |
1911 |
|
|
|
1912 |
root |
1.10 |
eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
1913 |
root |
1.6 |
{ |
1914 |
root |
1.9 |
REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1915 |
root |
1.6 |
} |
1916 |
|
|
|
1917 |
root |
1.10 |
eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
1918 |
root |
1.6 |
{ |
1919 |
root |
1.9 |
REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; |
1920 |
root |
1.6 |
} |
1921 |
|
|
|
1922 |
root |
1.10 |
eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data) |
1923 |
root |
1.6 |
{ |
1924 |
root |
1.9 |
REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; |
1925 |
root |
1.6 |
} |
1926 |
|
|
|
1927 |
root |
1.10 |
eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data) |
1928 |
root |
1.6 |
{ |
1929 |
root |
1.9 |
REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; |
1930 |
root |
1.6 |
} |
1931 |
|
|
|
1932 |
root |
1.10 |
eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data) |
1933 |
root |
1.6 |
{ |
1934 |
root |
1.9 |
REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; |
1935 |
root |
1.6 |
} |
1936 |
|
|
|
1937 |
root |
1.10 |
eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
1938 |
root |
1.6 |
{ |
1939 |
root |
1.9 |
REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
1940 |
root |
1.6 |
} |
1941 |
|
|
|
1942 |
root |
1.10 |
eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
1943 |
root |
1.6 |
{ |
1944 |
root |
1.9 |
REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1945 |
root |
1.6 |
} |
1946 |
|
|
|
1947 |
root |
1.10 |
eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
1948 |
root |
1.6 |
{ |
1949 |
root |
1.9 |
REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; |
1950 |
root |
1.6 |
} |
1951 |
|
|
|
1952 |
root |
1.10 |
eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
1953 |
root |
1.6 |
{ |
1954 |
root |
1.9 |
REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; |
1955 |
root |
1.6 |
} |
1956 |
|
|
|
1957 |
root |
1.9 |
static eio_req * |
1958 |
root |
1.10 |
eio__1path (int type, const char *path, int pri, eio_cb cb, void *data) |
1959 |
root |
1.6 |
{ |
1960 |
root |
1.9 |
REQ (type); PATH; SEND; |
1961 |
root |
1.6 |
} |
1962 |
|
|
|
1963 |
root |
1.10 |
eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1964 |
root |
1.6 |
{ |
1965 |
root |
1.10 |
return eio__1path (EIO_READLINK, path, pri, cb, data); |
1966 |
root |
1.6 |
} |
1967 |
|
|
|
1968 |
root |
1.10 |
eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1969 |
root |
1.6 |
{ |
1970 |
root |
1.10 |
return eio__1path (EIO_STAT, path, pri, cb, data); |
1971 |
root |
1.6 |
} |
1972 |
|
|
|
1973 |
root |
1.10 |
eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1974 |
root |
1.6 |
{ |
1975 |
root |
1.10 |
return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1976 |
root |
1.6 |
} |
1977 |
|
|
|
1978 |
root |
1.51 |
eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
1979 |
|
|
{ |
1980 |
|
|
return eio__1path (EIO_STATVFS, path, pri, cb, data); |
1981 |
|
|
} |
1982 |
|
|
|
1983 |
root |
1.10 |
eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1984 |
root |
1.6 |
{ |
1985 |
root |
1.10 |
return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1986 |
root |
1.6 |
} |
1987 |
|
|
|
1988 |
root |
1.10 |
eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1989 |
root |
1.6 |
{ |
1990 |
root |
1.10 |
return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1991 |
root |
1.6 |
} |
1992 |
|
|
|
1993 |
root |
1.31 |
eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1994 |
root |
1.1 |
{ |
1995 |
root |
1.31 |
REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1996 |
root |
1.1 |
} |
1997 |
|
|
|
1998 |
root |
1.10 |
eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1999 |
root |
1.1 |
{ |
2000 |
root |
1.61 |
REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
2001 |
root |
1.1 |
} |
2002 |
|
|
|
2003 |
root |
1.9 |
static eio_req * |
2004 |
root |
1.10 |
eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2005 |
root |
1.1 |
{ |
2006 |
root |
1.9 |
REQ (type); PATH; |
2007 |
root |
1.1 |
|
2008 |
root |
1.9 |
req->flags |= EIO_FLAG_PTR2_FREE; |
2009 |
|
|
req->ptr2 = strdup (new_path); |
2010 |
|
|
if (!req->ptr2) |
2011 |
|
|
{ |
2012 |
|
|
eio_api_destroy (req); |
2013 |
|
|
return 0; |
2014 |
|
|
} |
2015 |
root |
1.1 |
|
2016 |
root |
1.9 |
SEND; |
2017 |
root |
1.1 |
} |
2018 |
|
|
|
2019 |
root |
1.10 |
eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2020 |
root |
1.1 |
{ |
2021 |
root |
1.10 |
return eio__2path (EIO_LINK, path, new_path, pri, cb, data); |
2022 |
root |
1.1 |
} |
2023 |
|
|
|
2024 |
root |
1.10 |
eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2025 |
root |
1.1 |
{ |
2026 |
root |
1.10 |
return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data); |
2027 |
root |
1.1 |
} |
2028 |
|
|
|
2029 |
root |
1.10 |
eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2030 |
root |
1.1 |
{ |
2031 |
root |
1.10 |
return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2032 |
|
|
} |
2033 |
|
|
|
2034 |
root |
1.68 |
eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data); |
2035 |
root |
1.15 |
{ |
2036 |
root |
1.68 |
REQ (EIO_CUSTOM); req->feed = execute; SEND; |
2037 |
root |
1.15 |
} |
2038 |
|
|
|
2039 |
root |
1.12 |
#endif |
2040 |
|
|
|
2041 |
root |
1.10 |
eio_req *eio_grp (eio_cb cb, void *data) |
2042 |
|
|
{ |
2043 |
|
|
const int pri = EIO_PRI_MAX; |
2044 |
|
|
|
2045 |
|
|
REQ (EIO_GROUP); SEND; |
2046 |
root |
1.1 |
} |
2047 |
|
|
|
2048 |
root |
1.9 |
#undef REQ |
2049 |
|
|
#undef PATH |
2050 |
|
|
#undef SEND |
2051 |
root |
1.1 |
|
2052 |
root |
1.9 |
/*****************************************************************************/ |
2053 |
|
|
/* grp functions */ |
2054 |
root |
1.1 |
|
2055 |
root |
1.2 |
void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2056 |
root |
1.1 |
{ |
2057 |
|
|
grp->int2 = limit; |
2058 |
|
|
grp->feed = feed; |
2059 |
root |
1.2 |
|
2060 |
|
|
grp_try_feed (grp); |
2061 |
|
|
} |
2062 |
|
|
|
2063 |
|
|
void eio_grp_limit (eio_req *grp, int limit) |
2064 |
|
|
{ |
2065 |
|
|
grp->int2 = limit; |
2066 |
|
|
|
2067 |
|
|
grp_try_feed (grp); |
2068 |
root |
1.1 |
} |
2069 |
|
|
|
2070 |
|
|
void eio_grp_add (eio_req *grp, eio_req *req) |
2071 |
|
|
{ |
2072 |
|
|
assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2073 |
|
|
|
2074 |
root |
1.26 |
grp->flags |= EIO_FLAG_GROUPADD; |
2075 |
|
|
|
2076 |
root |
1.1 |
++grp->size; |
2077 |
|
|
req->grp = grp; |
2078 |
|
|
|
2079 |
|
|
req->grp_prev = 0; |
2080 |
|
|
req->grp_next = grp->grp_first; |
2081 |
|
|
|
2082 |
|
|
if (grp->grp_first) |
2083 |
|
|
grp->grp_first->grp_prev = req; |
2084 |
|
|
|
2085 |
|
|
grp->grp_first = req; |
2086 |
|
|
} |
2087 |
|
|
|
2088 |
root |
1.9 |
/*****************************************************************************/ |
2089 |
|
|
/* misc garbage */ |
2090 |
|
|
|
2091 |
|
|
ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2092 |
|
|
{ |
2093 |
root |
1.14 |
etp_worker wrk; |
2094 |
root |
1.37 |
ssize_t ret; |
2095 |
root |
1.9 |
|
2096 |
|
|
wrk.dbuf = 0; |
2097 |
|
|
|
2098 |
root |
1.37 |
ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
2099 |
root |
1.9 |
|
2100 |
|
|
if (wrk.dbuf) |
2101 |
|
|
free (wrk.dbuf); |
2102 |
root |
1.37 |
|
2103 |
|
|
return ret; |
2104 |
root |
1.9 |
} |
2105 |
root |
1.1 |
|