| 1 |
/* |
| 2 |
* libeio implementation |
| 3 |
* |
| 4 |
* Copyright (c) 2007,2008,2009 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 |
#include "eio.h" |
| 41 |
|
| 42 |
#ifdef EIO_STACKSIZE |
| 43 |
# define XTHREAD_STACKSIZE EIO_STACKSIZE |
| 44 |
#endif |
| 45 |
#include "xthread.h" |
| 46 |
|
| 47 |
#include <errno.h> |
| 48 |
#include <stddef.h> |
| 49 |
#include <stdlib.h> |
| 50 |
#include <string.h> |
| 51 |
#include <errno.h> |
| 52 |
#include <sys/types.h> |
| 53 |
#include <sys/stat.h> |
| 54 |
#include <limits.h> |
| 55 |
#include <fcntl.h> |
| 56 |
#include <assert.h> |
| 57 |
|
| 58 |
#ifndef EIO_FINISH |
| 59 |
# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
| 60 |
#endif |
| 61 |
|
| 62 |
#ifndef EIO_DESTROY |
| 63 |
# define EIO_DESTROY(req) do { if ((req)->destroy) (req)->destroy (req); } while (0) |
| 64 |
#endif |
| 65 |
|
| 66 |
#ifndef EIO_FEED |
| 67 |
# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
| 68 |
#endif |
| 69 |
|
| 70 |
#ifdef _WIN32 |
| 71 |
|
| 72 |
/*doh*/ |
| 73 |
#else |
| 74 |
|
| 75 |
# include "config.h" |
| 76 |
# include <sys/time.h> |
| 77 |
# include <sys/select.h> |
| 78 |
# include <sys/mman.h> |
| 79 |
# include <unistd.h> |
| 80 |
# include <utime.h> |
| 81 |
# include <signal.h> |
| 82 |
# include <dirent.h> |
| 83 |
|
| 84 |
/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
| 85 |
# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
| 86 |
# define _DIRENT_HAVE_D_TYPE /* sigh */ |
| 87 |
# define D_INO(de) (de)->d_fileno |
| 88 |
# define D_NAMLEN(de) (de)->d_namlen |
| 89 |
# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
| 90 |
# define D_INO(de) (de)->d_ino |
| 91 |
# endif |
| 92 |
|
| 93 |
#ifdef _D_EXACT_NAMLEN |
| 94 |
# undef D_NAMLEN |
| 95 |
# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
| 96 |
#endif |
| 97 |
|
| 98 |
# ifdef _DIRENT_HAVE_D_TYPE |
| 99 |
# define D_TYPE(de) (de)->d_type |
| 100 |
# endif |
| 101 |
|
| 102 |
# ifndef EIO_STRUCT_DIRENT |
| 103 |
# define EIO_STRUCT_DIRENT struct dirent |
| 104 |
# endif |
| 105 |
|
| 106 |
#endif |
| 107 |
|
| 108 |
#if HAVE_SENDFILE |
| 109 |
# if __linux |
| 110 |
# include <sys/sendfile.h> |
| 111 |
# elif __freebsd |
| 112 |
# include <sys/socket.h> |
| 113 |
# include <sys/uio.h> |
| 114 |
# elif __hpux |
| 115 |
# include <sys/socket.h> |
| 116 |
# elif __solaris /* not yet */ |
| 117 |
# include <sys/sendfile.h> |
| 118 |
# else |
| 119 |
# error sendfile support requested but not available |
| 120 |
# endif |
| 121 |
#endif |
| 122 |
|
| 123 |
#ifndef D_TYPE |
| 124 |
# define D_TYPE(de) 0 |
| 125 |
#endif |
| 126 |
#ifndef D_INO |
| 127 |
# define D_INO(de) 0 |
| 128 |
#endif |
| 129 |
#ifndef D_NAMLEN |
| 130 |
# define D_NAMLEN(de) strlen ((de)->d_name) |
| 131 |
#endif |
| 132 |
|
| 133 |
/* number of seconds after which an idle threads exit */ |
| 134 |
#define IDLE_TIMEOUT 10 |
| 135 |
|
| 136 |
/* used for struct dirent, AIX doesn't provide it */ |
| 137 |
#ifndef NAME_MAX |
| 138 |
# define NAME_MAX 4096 |
| 139 |
#endif |
| 140 |
|
| 141 |
/* buffer size for various temporary buffers */ |
| 142 |
#define EIO_BUFSIZE 65536 |
| 143 |
|
| 144 |
#define dBUF \ |
| 145 |
char *eio_buf; \ |
| 146 |
ETP_WORKER_LOCK (self); \ |
| 147 |
self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
| 148 |
ETP_WORKER_UNLOCK (self); \ |
| 149 |
errno = ENOMEM; \ |
| 150 |
if (!eio_buf) \ |
| 151 |
return -1; |
| 152 |
|
| 153 |
#define EIO_TICKS ((1000000 + 1023) >> 10) |
| 154 |
|
| 155 |
/*****************************************************************************/ |
| 156 |
|
| 157 |
#if __GNUC__ >= 3 |
| 158 |
# define expect(expr,value) __builtin_expect ((expr),(value)) |
| 159 |
#else |
| 160 |
# define expect(expr,value) (expr) |
| 161 |
#endif |
| 162 |
|
| 163 |
#define expect_false(expr) expect ((expr) != 0, 0) |
| 164 |
#define expect_true(expr) expect ((expr) != 0, 1) |
| 165 |
|
| 166 |
/*****************************************************************************/ |
| 167 |
|
| 168 |
#define ETP_PRI_MIN EIO_PRI_MIN |
| 169 |
#define ETP_PRI_MAX EIO_PRI_MAX |
| 170 |
|
| 171 |
struct etp_worker; |
| 172 |
|
| 173 |
#define ETP_REQ eio_req |
| 174 |
#define ETP_DESTROY(req) eio_destroy (req) |
| 175 |
static int eio_finish (eio_req *req); |
| 176 |
#define ETP_FINISH(req) eio_finish (req) |
| 177 |
static void eio_execute (struct etp_worker *self, eio_req *req); |
| 178 |
#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
| 179 |
|
| 180 |
#define ETP_WORKER_CLEAR(req) \ |
| 181 |
if (wrk->dbuf) \ |
| 182 |
{ \ |
| 183 |
free (wrk->dbuf); \ |
| 184 |
wrk->dbuf = 0; \ |
| 185 |
} \ |
| 186 |
\ |
| 187 |
if (wrk->dirp) \ |
| 188 |
{ \ |
| 189 |
closedir (wrk->dirp); \ |
| 190 |
wrk->dirp = 0; \ |
| 191 |
} |
| 192 |
|
| 193 |
#define ETP_WORKER_COMMON \ |
| 194 |
void *dbuf; \ |
| 195 |
DIR *dirp; |
| 196 |
|
| 197 |
/*****************************************************************************/ |
| 198 |
|
| 199 |
#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
| 200 |
|
| 201 |
/* calculcate time difference in ~1/EIO_TICKS of a second */ |
| 202 |
static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
| 203 |
{ |
| 204 |
return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
| 205 |
+ ((tv2->tv_usec - tv1->tv_usec) >> 10); |
| 206 |
} |
| 207 |
|
| 208 |
static unsigned int started, idle, wanted = 4; |
| 209 |
|
| 210 |
static void (*want_poll_cb) (void); |
| 211 |
static void (*done_poll_cb) (void); |
| 212 |
|
| 213 |
static unsigned int max_poll_time; /* reslock */ |
| 214 |
static unsigned int max_poll_reqs; /* reslock */ |
| 215 |
|
| 216 |
static volatile unsigned int nreqs; /* reqlock */ |
| 217 |
static volatile unsigned int nready; /* reqlock */ |
| 218 |
static volatile unsigned int npending; /* reqlock */ |
| 219 |
static volatile unsigned int max_idle = 4; |
| 220 |
|
| 221 |
static mutex_t wrklock = X_MUTEX_INIT; |
| 222 |
static mutex_t reslock = X_MUTEX_INIT; |
| 223 |
static mutex_t reqlock = X_MUTEX_INIT; |
| 224 |
static cond_t reqwait = X_COND_INIT; |
| 225 |
|
| 226 |
#if !HAVE_PREADWRITE |
| 227 |
/* |
| 228 |
* make our pread/pwrite emulation safe against themselves, but not against |
| 229 |
* normal read/write by using a mutex. slows down execution a lot, |
| 230 |
* but that's your problem, not mine. |
| 231 |
*/ |
| 232 |
static mutex_t preadwritelock = X_MUTEX_INIT; |
| 233 |
#endif |
| 234 |
|
| 235 |
typedef struct etp_worker |
| 236 |
{ |
| 237 |
/* locked by wrklock */ |
| 238 |
struct etp_worker *prev, *next; |
| 239 |
|
| 240 |
thread_t tid; |
| 241 |
|
| 242 |
/* locked by reslock, reqlock or wrklock */ |
| 243 |
ETP_REQ *req; /* currently processed request */ |
| 244 |
|
| 245 |
ETP_WORKER_COMMON |
| 246 |
} etp_worker; |
| 247 |
|
| 248 |
static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
| 249 |
|
| 250 |
#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
| 251 |
#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
| 252 |
|
| 253 |
/* worker threads management */ |
| 254 |
|
| 255 |
static void etp_worker_clear (etp_worker *wrk) |
| 256 |
{ |
| 257 |
ETP_WORKER_CLEAR (wrk); |
| 258 |
} |
| 259 |
|
| 260 |
static void etp_worker_free (etp_worker *wrk) |
| 261 |
{ |
| 262 |
wrk->next->prev = wrk->prev; |
| 263 |
wrk->prev->next = wrk->next; |
| 264 |
|
| 265 |
free (wrk); |
| 266 |
} |
| 267 |
|
| 268 |
static unsigned int etp_nreqs (void) |
| 269 |
{ |
| 270 |
int retval; |
| 271 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
| 272 |
retval = nreqs; |
| 273 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
| 274 |
return retval; |
| 275 |
} |
| 276 |
|
| 277 |
static unsigned int etp_nready (void) |
| 278 |
{ |
| 279 |
unsigned int retval; |
| 280 |
|
| 281 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
| 282 |
retval = nready; |
| 283 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
| 284 |
|
| 285 |
return retval; |
| 286 |
} |
| 287 |
|
| 288 |
static unsigned int etp_npending (void) |
| 289 |
{ |
| 290 |
unsigned int retval; |
| 291 |
|
| 292 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
| 293 |
retval = npending; |
| 294 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
| 295 |
|
| 296 |
return retval; |
| 297 |
} |
| 298 |
|
| 299 |
static unsigned int etp_nthreads (void) |
| 300 |
{ |
| 301 |
unsigned int retval; |
| 302 |
|
| 303 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
| 304 |
retval = started; |
| 305 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
| 306 |
|
| 307 |
return retval; |
| 308 |
} |
| 309 |
|
| 310 |
/* |
| 311 |
* a somewhat faster data structure might be nice, but |
| 312 |
* with 8 priorities this actually needs <20 insns |
| 313 |
* per shift, the most expensive operation. |
| 314 |
*/ |
| 315 |
typedef struct { |
| 316 |
ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */ |
| 317 |
int size; |
| 318 |
} etp_reqq; |
| 319 |
|
| 320 |
static etp_reqq req_queue; |
| 321 |
static etp_reqq res_queue; |
| 322 |
|
| 323 |
static int reqq_push (etp_reqq *q, ETP_REQ *req) |
| 324 |
{ |
| 325 |
int pri = req->pri; |
| 326 |
req->next = 0; |
| 327 |
|
| 328 |
if (q->qe[pri]) |
| 329 |
{ |
| 330 |
q->qe[pri]->next = req; |
| 331 |
q->qe[pri] = req; |
| 332 |
} |
| 333 |
else |
| 334 |
q->qe[pri] = q->qs[pri] = req; |
| 335 |
|
| 336 |
return q->size++; |
| 337 |
} |
| 338 |
|
| 339 |
static ETP_REQ *reqq_shift (etp_reqq *q) |
| 340 |
{ |
| 341 |
int pri; |
| 342 |
|
| 343 |
if (!q->size) |
| 344 |
return 0; |
| 345 |
|
| 346 |
--q->size; |
| 347 |
|
| 348 |
for (pri = ETP_NUM_PRI; pri--; ) |
| 349 |
{ |
| 350 |
eio_req *req = q->qs[pri]; |
| 351 |
|
| 352 |
if (req) |
| 353 |
{ |
| 354 |
if (!(q->qs[pri] = (eio_req *)req->next)) |
| 355 |
q->qe[pri] = 0; |
| 356 |
|
| 357 |
return req; |
| 358 |
} |
| 359 |
} |
| 360 |
|
| 361 |
abort (); |
| 362 |
} |
| 363 |
|
| 364 |
static void etp_atfork_prepare (void) |
| 365 |
{ |
| 366 |
X_LOCK (wrklock); |
| 367 |
X_LOCK (reqlock); |
| 368 |
X_LOCK (reslock); |
| 369 |
#if !HAVE_PREADWRITE |
| 370 |
X_LOCK (preadwritelock); |
| 371 |
#endif |
| 372 |
} |
| 373 |
|
| 374 |
static void etp_atfork_parent (void) |
| 375 |
{ |
| 376 |
#if !HAVE_PREADWRITE |
| 377 |
X_UNLOCK (preadwritelock); |
| 378 |
#endif |
| 379 |
X_UNLOCK (reslock); |
| 380 |
X_UNLOCK (reqlock); |
| 381 |
X_UNLOCK (wrklock); |
| 382 |
} |
| 383 |
|
| 384 |
static void etp_atfork_child (void) |
| 385 |
{ |
| 386 |
ETP_REQ *prv; |
| 387 |
|
| 388 |
while ((prv = reqq_shift (&req_queue))) |
| 389 |
ETP_DESTROY (prv); |
| 390 |
|
| 391 |
while ((prv = reqq_shift (&res_queue))) |
| 392 |
ETP_DESTROY (prv); |
| 393 |
|
| 394 |
while (wrk_first.next != &wrk_first) |
| 395 |
{ |
| 396 |
etp_worker *wrk = wrk_first.next; |
| 397 |
|
| 398 |
if (wrk->req) |
| 399 |
ETP_DESTROY (wrk->req); |
| 400 |
|
| 401 |
etp_worker_clear (wrk); |
| 402 |
etp_worker_free (wrk); |
| 403 |
} |
| 404 |
|
| 405 |
started = 0; |
| 406 |
idle = 0; |
| 407 |
nreqs = 0; |
| 408 |
nready = 0; |
| 409 |
npending = 0; |
| 410 |
|
| 411 |
etp_atfork_parent (); |
| 412 |
} |
| 413 |
|
| 414 |
static void |
| 415 |
etp_once_init (void) |
| 416 |
{ |
| 417 |
X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
| 418 |
} |
| 419 |
|
| 420 |
static int |
| 421 |
etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
| 422 |
{ |
| 423 |
static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
| 424 |
|
| 425 |
pthread_once (&doinit, etp_once_init); |
| 426 |
|
| 427 |
want_poll_cb = want_poll; |
| 428 |
done_poll_cb = done_poll; |
| 429 |
|
| 430 |
return 0; |
| 431 |
} |
| 432 |
|
| 433 |
X_THREAD_PROC (etp_proc); |
| 434 |
|
| 435 |
static void etp_start_thread (void) |
| 436 |
{ |
| 437 |
etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
| 438 |
|
| 439 |
/*TODO*/ |
| 440 |
assert (("unable to allocate worker thread data", wrk)); |
| 441 |
|
| 442 |
X_LOCK (wrklock); |
| 443 |
|
| 444 |
if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) |
| 445 |
{ |
| 446 |
wrk->prev = &wrk_first; |
| 447 |
wrk->next = wrk_first.next; |
| 448 |
wrk_first.next->prev = wrk; |
| 449 |
wrk_first.next = wrk; |
| 450 |
++started; |
| 451 |
} |
| 452 |
else |
| 453 |
free (wrk); |
| 454 |
|
| 455 |
X_UNLOCK (wrklock); |
| 456 |
} |
| 457 |
|
| 458 |
static void etp_maybe_start_thread (void) |
| 459 |
{ |
| 460 |
if (expect_true (etp_nthreads () >= wanted)) |
| 461 |
return; |
| 462 |
|
| 463 |
/* todo: maybe use idle here, but might be less exact */ |
| 464 |
if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
| 465 |
return; |
| 466 |
|
| 467 |
etp_start_thread (); |
| 468 |
} |
| 469 |
|
| 470 |
static void etp_end_thread (void) |
| 471 |
{ |
| 472 |
eio_req *req = calloc (1, sizeof (eio_req)); |
| 473 |
|
| 474 |
req->type = -1; |
| 475 |
req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
| 476 |
|
| 477 |
X_LOCK (reqlock); |
| 478 |
reqq_push (&req_queue, req); |
| 479 |
X_COND_SIGNAL (reqwait); |
| 480 |
X_UNLOCK (reqlock); |
| 481 |
|
| 482 |
X_LOCK (wrklock); |
| 483 |
--started; |
| 484 |
X_UNLOCK (wrklock); |
| 485 |
} |
| 486 |
|
| 487 |
static int etp_poll (void) |
| 488 |
{ |
| 489 |
unsigned int maxreqs; |
| 490 |
unsigned int maxtime; |
| 491 |
struct timeval tv_start, tv_now; |
| 492 |
|
| 493 |
X_LOCK (reslock); |
| 494 |
maxreqs = max_poll_reqs; |
| 495 |
maxtime = max_poll_time; |
| 496 |
X_UNLOCK (reslock); |
| 497 |
|
| 498 |
if (maxtime) |
| 499 |
gettimeofday (&tv_start, 0); |
| 500 |
|
| 501 |
for (;;) |
| 502 |
{ |
| 503 |
ETP_REQ *req; |
| 504 |
|
| 505 |
etp_maybe_start_thread (); |
| 506 |
|
| 507 |
X_LOCK (reslock); |
| 508 |
req = reqq_shift (&res_queue); |
| 509 |
|
| 510 |
if (req) |
| 511 |
{ |
| 512 |
--npending; |
| 513 |
|
| 514 |
if (!res_queue.size && done_poll_cb) |
| 515 |
done_poll_cb (); |
| 516 |
} |
| 517 |
|
| 518 |
X_UNLOCK (reslock); |
| 519 |
|
| 520 |
if (!req) |
| 521 |
return 0; |
| 522 |
|
| 523 |
X_LOCK (reqlock); |
| 524 |
--nreqs; |
| 525 |
X_UNLOCK (reqlock); |
| 526 |
|
| 527 |
if (expect_false (req->type == EIO_GROUP && req->size)) |
| 528 |
{ |
| 529 |
req->int1 = 1; /* mark request as delayed */ |
| 530 |
continue; |
| 531 |
} |
| 532 |
else |
| 533 |
{ |
| 534 |
int res = ETP_FINISH (req); |
| 535 |
if (expect_false (res)) |
| 536 |
return res; |
| 537 |
} |
| 538 |
|
| 539 |
if (expect_false (maxreqs && !--maxreqs)) |
| 540 |
break; |
| 541 |
|
| 542 |
if (maxtime) |
| 543 |
{ |
| 544 |
gettimeofday (&tv_now, 0); |
| 545 |
|
| 546 |
if (tvdiff (&tv_start, &tv_now) >= maxtime) |
| 547 |
break; |
| 548 |
} |
| 549 |
} |
| 550 |
|
| 551 |
errno = EAGAIN; |
| 552 |
return -1; |
| 553 |
} |
| 554 |
|
| 555 |
static void etp_cancel (ETP_REQ *req) |
| 556 |
{ |
| 557 |
X_LOCK (wrklock); |
| 558 |
req->flags |= EIO_FLAG_CANCELLED; |
| 559 |
X_UNLOCK (wrklock); |
| 560 |
|
| 561 |
eio_grp_cancel (req); |
| 562 |
} |
| 563 |
|
| 564 |
static void etp_submit (ETP_REQ *req) |
| 565 |
{ |
| 566 |
req->pri -= ETP_PRI_MIN; |
| 567 |
|
| 568 |
if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
| 569 |
if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
| 570 |
|
| 571 |
if (expect_false (req->type == EIO_GROUP)) |
| 572 |
{ |
| 573 |
/* I hope this is worth it :/ */ |
| 574 |
X_LOCK (reqlock); |
| 575 |
++nreqs; |
| 576 |
X_UNLOCK (reqlock); |
| 577 |
|
| 578 |
X_LOCK (reslock); |
| 579 |
|
| 580 |
++npending; |
| 581 |
|
| 582 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
| 583 |
want_poll_cb (); |
| 584 |
|
| 585 |
X_UNLOCK (reslock); |
| 586 |
} |
| 587 |
else |
| 588 |
{ |
| 589 |
X_LOCK (reqlock); |
| 590 |
++nreqs; |
| 591 |
++nready; |
| 592 |
reqq_push (&req_queue, req); |
| 593 |
X_COND_SIGNAL (reqwait); |
| 594 |
X_UNLOCK (reqlock); |
| 595 |
|
| 596 |
etp_maybe_start_thread (); |
| 597 |
} |
| 598 |
} |
| 599 |
|
| 600 |
static void etp_set_max_poll_time (double nseconds) |
| 601 |
{ |
| 602 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
| 603 |
max_poll_time = nseconds; |
| 604 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
| 605 |
} |
| 606 |
|
| 607 |
static void etp_set_max_poll_reqs (unsigned int maxreqs) |
| 608 |
{ |
| 609 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
| 610 |
max_poll_reqs = maxreqs; |
| 611 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
| 612 |
} |
| 613 |
|
| 614 |
static void etp_set_max_idle (unsigned int nthreads) |
| 615 |
{ |
| 616 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
| 617 |
max_idle = nthreads <= 0 ? 1 : nthreads; |
| 618 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
| 619 |
} |
| 620 |
|
| 621 |
static void etp_set_min_parallel (unsigned int nthreads) |
| 622 |
{ |
| 623 |
if (wanted < nthreads) |
| 624 |
wanted = nthreads; |
| 625 |
} |
| 626 |
|
| 627 |
static void etp_set_max_parallel (unsigned int nthreads) |
| 628 |
{ |
| 629 |
if (wanted > nthreads) |
| 630 |
wanted = nthreads; |
| 631 |
|
| 632 |
while (started > wanted) |
| 633 |
etp_end_thread (); |
| 634 |
} |
| 635 |
|
| 636 |
/*****************************************************************************/ |
| 637 |
|
| 638 |
static void grp_try_feed (eio_req *grp) |
| 639 |
{ |
| 640 |
while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
| 641 |
{ |
| 642 |
grp->flags &= ~EIO_FLAG_GROUPADD; |
| 643 |
|
| 644 |
EIO_FEED (grp); |
| 645 |
|
| 646 |
/* stop if no progress has been made */ |
| 647 |
if (!(grp->flags & EIO_FLAG_GROUPADD)) |
| 648 |
{ |
| 649 |
grp->feed = 0; |
| 650 |
break; |
| 651 |
} |
| 652 |
} |
| 653 |
} |
| 654 |
|
| 655 |
static int grp_dec (eio_req *grp) |
| 656 |
{ |
| 657 |
--grp->size; |
| 658 |
|
| 659 |
/* call feeder, if applicable */ |
| 660 |
grp_try_feed (grp); |
| 661 |
|
| 662 |
/* finish, if done */ |
| 663 |
if (!grp->size && grp->int1) |
| 664 |
return eio_finish (grp); |
| 665 |
else |
| 666 |
return 0; |
| 667 |
} |
| 668 |
|
| 669 |
void eio_destroy (eio_req *req) |
| 670 |
{ |
| 671 |
if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
| 672 |
if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
| 673 |
|
| 674 |
EIO_DESTROY (req); |
| 675 |
} |
| 676 |
|
| 677 |
static int eio_finish (eio_req *req) |
| 678 |
{ |
| 679 |
int res = EIO_FINISH (req); |
| 680 |
|
| 681 |
if (req->grp) |
| 682 |
{ |
| 683 |
int res2; |
| 684 |
eio_req *grp = req->grp; |
| 685 |
|
| 686 |
/* unlink request */ |
| 687 |
if (req->grp_next) req->grp_next->grp_prev = req->grp_prev; |
| 688 |
if (req->grp_prev) req->grp_prev->grp_next = req->grp_next; |
| 689 |
|
| 690 |
if (grp->grp_first == req) |
| 691 |
grp->grp_first = req->grp_next; |
| 692 |
|
| 693 |
res2 = grp_dec (grp); |
| 694 |
|
| 695 |
if (!res && res2) |
| 696 |
res = res2; |
| 697 |
} |
| 698 |
|
| 699 |
eio_destroy (req); |
| 700 |
|
| 701 |
return res; |
| 702 |
} |
| 703 |
|
| 704 |
void eio_grp_cancel (eio_req *grp) |
| 705 |
{ |
| 706 |
for (grp = grp->grp_first; grp; grp = grp->grp_next) |
| 707 |
eio_cancel (grp); |
| 708 |
} |
| 709 |
|
| 710 |
void eio_cancel (eio_req *req) |
| 711 |
{ |
| 712 |
etp_cancel (req); |
| 713 |
} |
| 714 |
|
| 715 |
void eio_submit (eio_req *req) |
| 716 |
{ |
| 717 |
etp_submit (req); |
| 718 |
} |
| 719 |
|
| 720 |
unsigned int eio_nreqs (void) |
| 721 |
{ |
| 722 |
return etp_nreqs (); |
| 723 |
} |
| 724 |
|
| 725 |
unsigned int eio_nready (void) |
| 726 |
{ |
| 727 |
return etp_nready (); |
| 728 |
} |
| 729 |
|
| 730 |
unsigned int eio_npending (void) |
| 731 |
{ |
| 732 |
return etp_npending (); |
| 733 |
} |
| 734 |
|
| 735 |
unsigned int eio_nthreads (void) |
| 736 |
{ |
| 737 |
return etp_nthreads (); |
| 738 |
} |
| 739 |
|
| 740 |
void eio_set_max_poll_time (double nseconds) |
| 741 |
{ |
| 742 |
etp_set_max_poll_time (nseconds); |
| 743 |
} |
| 744 |
|
| 745 |
void eio_set_max_poll_reqs (unsigned int maxreqs) |
| 746 |
{ |
| 747 |
etp_set_max_poll_reqs (maxreqs); |
| 748 |
} |
| 749 |
|
| 750 |
void eio_set_max_idle (unsigned int nthreads) |
| 751 |
{ |
| 752 |
etp_set_max_idle (nthreads); |
| 753 |
} |
| 754 |
|
| 755 |
void eio_set_min_parallel (unsigned int nthreads) |
| 756 |
{ |
| 757 |
etp_set_min_parallel (nthreads); |
| 758 |
} |
| 759 |
|
| 760 |
void eio_set_max_parallel (unsigned int nthreads) |
| 761 |
{ |
| 762 |
etp_set_max_parallel (nthreads); |
| 763 |
} |
| 764 |
|
| 765 |
int eio_poll (void) |
| 766 |
{ |
| 767 |
return etp_poll (); |
| 768 |
} |
| 769 |
|
| 770 |
/*****************************************************************************/ |
| 771 |
/* work around various missing functions */ |
| 772 |
|
| 773 |
#if !HAVE_PREADWRITE |
| 774 |
# undef pread |
| 775 |
# undef pwrite |
| 776 |
# define pread eio__pread |
| 777 |
# define pwrite eio__pwrite |
| 778 |
|
| 779 |
static ssize_t |
| 780 |
eio__pread (int fd, void *buf, size_t count, off_t offset) |
| 781 |
{ |
| 782 |
ssize_t res; |
| 783 |
off_t ooffset; |
| 784 |
|
| 785 |
X_LOCK (preadwritelock); |
| 786 |
ooffset = lseek (fd, 0, SEEK_CUR); |
| 787 |
lseek (fd, offset, SEEK_SET); |
| 788 |
res = read (fd, buf, count); |
| 789 |
lseek (fd, ooffset, SEEK_SET); |
| 790 |
X_UNLOCK (preadwritelock); |
| 791 |
|
| 792 |
return res; |
| 793 |
} |
| 794 |
|
| 795 |
static ssize_t |
| 796 |
eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
| 797 |
{ |
| 798 |
ssize_t res; |
| 799 |
off_t ooffset; |
| 800 |
|
| 801 |
X_LOCK (preadwritelock); |
| 802 |
ooffset = lseek (fd, 0, SEEK_CUR); |
| 803 |
lseek (fd, offset, SEEK_SET); |
| 804 |
res = write (fd, buf, count); |
| 805 |
lseek (fd, ooffset, SEEK_SET); |
| 806 |
X_UNLOCK (preadwritelock); |
| 807 |
|
| 808 |
return res; |
| 809 |
} |
| 810 |
#endif |
| 811 |
|
| 812 |
#ifndef HAVE_FUTIMES |
| 813 |
|
| 814 |
# undef utimes |
| 815 |
# undef futimes |
| 816 |
# define utimes(path,times) eio__utimes (path, times) |
| 817 |
# define futimes(fd,times) eio__futimes (fd, times) |
| 818 |
|
| 819 |
static int |
| 820 |
eio__utimes (const char *filename, const struct timeval times[2]) |
| 821 |
{ |
| 822 |
if (times) |
| 823 |
{ |
| 824 |
struct utimbuf buf; |
| 825 |
|
| 826 |
buf.actime = times[0].tv_sec; |
| 827 |
buf.modtime = times[1].tv_sec; |
| 828 |
|
| 829 |
return utime (filename, &buf); |
| 830 |
} |
| 831 |
else |
| 832 |
return utime (filename, 0); |
| 833 |
} |
| 834 |
|
| 835 |
static int eio__futimes (int fd, const struct timeval tv[2]) |
| 836 |
{ |
| 837 |
errno = ENOSYS; |
| 838 |
return -1; |
| 839 |
} |
| 840 |
|
| 841 |
#endif |
| 842 |
|
| 843 |
#if !HAVE_FDATASYNC |
| 844 |
# undef fdatasync |
| 845 |
# define fdatasync(fd) fsync (fd) |
| 846 |
#endif |
| 847 |
|
| 848 |
/* sync_file_range always needs emulation */ |
| 849 |
int |
| 850 |
eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
| 851 |
{ |
| 852 |
#if HAVE_SYNC_FILE_RANGE |
| 853 |
int res; |
| 854 |
|
| 855 |
if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
| 856 |
|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
| 857 |
|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
| 858 |
{ |
| 859 |
flags = 0 |
| 860 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
| 861 |
| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
| 862 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
| 863 |
} |
| 864 |
|
| 865 |
res = sync_file_range (fd, offset, nbytes, flags); |
| 866 |
|
| 867 |
if (!res || errno != ENOSYS) |
| 868 |
return res; |
| 869 |
#endif |
| 870 |
|
| 871 |
/* even though we could play tricks with the flags, it's better to always |
| 872 |
* call fdatasync, as thta matches the expectation of it's users best */ |
| 873 |
return fdatasync (fd); |
| 874 |
} |
| 875 |
|
| 876 |
#if !HAVE_READAHEAD |
| 877 |
# undef readahead |
| 878 |
# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
| 879 |
|
| 880 |
static ssize_t |
| 881 |
eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
| 882 |
{ |
| 883 |
size_t todo = count; |
| 884 |
dBUF; |
| 885 |
|
| 886 |
while (todo > 0) |
| 887 |
{ |
| 888 |
size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; |
| 889 |
|
| 890 |
pread (fd, eio_buf, len, offset); |
| 891 |
offset += len; |
| 892 |
todo -= len; |
| 893 |
} |
| 894 |
|
| 895 |
errno = 0; |
| 896 |
return count; |
| 897 |
} |
| 898 |
|
| 899 |
#endif |
| 900 |
|
| 901 |
/* sendfile always needs emulation */ |
| 902 |
static ssize_t |
| 903 |
eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
| 904 |
{ |
| 905 |
ssize_t res; |
| 906 |
|
| 907 |
if (!count) |
| 908 |
return 0; |
| 909 |
|
| 910 |
#if HAVE_SENDFILE |
| 911 |
# if __linux |
| 912 |
res = sendfile (ofd, ifd, &offset, count); |
| 913 |
|
| 914 |
# elif __freebsd |
| 915 |
/* |
| 916 |
* Of course, the freebsd sendfile is a dire hack with no thoughts |
| 917 |
* wasted on making it similar to other I/O functions. |
| 918 |
*/ |
| 919 |
{ |
| 920 |
off_t sbytes; |
| 921 |
res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
| 922 |
|
| 923 |
if (res < 0 && sbytes) |
| 924 |
/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
| 925 |
res = sbytes; |
| 926 |
} |
| 927 |
|
| 928 |
# elif __hpux |
| 929 |
res = sendfile (ofd, ifd, offset, count, 0, 0); |
| 930 |
|
| 931 |
# elif __solaris |
| 932 |
{ |
| 933 |
struct sendfilevec vec; |
| 934 |
size_t sbytes; |
| 935 |
|
| 936 |
vec.sfv_fd = ifd; |
| 937 |
vec.sfv_flag = 0; |
| 938 |
vec.sfv_off = offset; |
| 939 |
vec.sfv_len = count; |
| 940 |
|
| 941 |
res = sendfilev (ofd, &vec, 1, &sbytes); |
| 942 |
|
| 943 |
if (res < 0 && sbytes) |
| 944 |
res = sbytes; |
| 945 |
} |
| 946 |
|
| 947 |
# endif |
| 948 |
#else |
| 949 |
res = -1; |
| 950 |
errno = ENOSYS; |
| 951 |
#endif |
| 952 |
|
| 953 |
if (res < 0 |
| 954 |
&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
| 955 |
#if __solaris |
| 956 |
|| errno == EAFNOSUPPORT || errno == EPROTOTYPE |
| 957 |
#endif |
| 958 |
) |
| 959 |
) |
| 960 |
{ |
| 961 |
/* emulate sendfile. this is a major pain in the ass */ |
| 962 |
dBUF; |
| 963 |
|
| 964 |
res = 0; |
| 965 |
|
| 966 |
while (count) |
| 967 |
{ |
| 968 |
ssize_t cnt; |
| 969 |
|
| 970 |
cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
| 971 |
|
| 972 |
if (cnt <= 0) |
| 973 |
{ |
| 974 |
if (cnt && !res) res = -1; |
| 975 |
break; |
| 976 |
} |
| 977 |
|
| 978 |
cnt = write (ofd, eio_buf, cnt); |
| 979 |
|
| 980 |
if (cnt <= 0) |
| 981 |
{ |
| 982 |
if (cnt && !res) res = -1; |
| 983 |
break; |
| 984 |
} |
| 985 |
|
| 986 |
offset += cnt; |
| 987 |
res += cnt; |
| 988 |
count -= cnt; |
| 989 |
} |
| 990 |
} |
| 991 |
|
| 992 |
return res; |
| 993 |
} |
| 994 |
|
| 995 |
static signed char |
| 996 |
eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
| 997 |
{ |
| 998 |
return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
| 999 |
: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
| 1000 |
} |
| 1001 |
|
| 1002 |
#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
| 1003 |
#define EIO_DENT_NUM_SCORES 9 |
| 1004 |
|
| 1005 |
#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
| 1006 |
#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ |
| 1007 |
|
| 1008 |
static void |
| 1009 |
eio_dent_sort (eio_dirent *dents, int size) |
| 1010 |
{ |
| 1011 |
if (size <= 1) |
| 1012 |
return; /* our insertion sort relies on size > 0 */ |
| 1013 |
|
| 1014 |
if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ |
| 1015 |
{ |
| 1016 |
/* first, use quicksort */ |
| 1017 |
/* should be good for 2**31 entries */ |
| 1018 |
int i; |
| 1019 |
struct rng { eio_dirent *l, *r; } rng [32]; |
| 1020 |
|
| 1021 |
i = 0; |
| 1022 |
rng[0].l = dents; |
| 1023 |
rng[0].r = dents + size; |
| 1024 |
|
| 1025 |
while (expect_true (i >= 0)) |
| 1026 |
{ |
| 1027 |
eio_dirent *L = rng [i].l; |
| 1028 |
eio_dirent *R = rng [i].r - 1; |
| 1029 |
|
| 1030 |
if (expect_false (L + EIO_QSORT_CUTOFF < R)) |
| 1031 |
{ |
| 1032 |
eio_dirent *mid = &dents [((L - dents) + (R - dents)) >> 1]; |
| 1033 |
eio_dirent piv = *mid; *mid = *L; *L = piv; |
| 1034 |
|
| 1035 |
while (L < R) |
| 1036 |
{ |
| 1037 |
while (EIO_DENT_CMP (*R, >=, piv) && L < R) |
| 1038 |
--R; |
| 1039 |
|
| 1040 |
if (L < R) |
| 1041 |
*L++ = *R; |
| 1042 |
|
| 1043 |
while (EIO_DENT_CMP (*L, <=, piv) && L < R) |
| 1044 |
++L; |
| 1045 |
|
| 1046 |
if (L < R) |
| 1047 |
*R-- = *L; |
| 1048 |
} |
| 1049 |
|
| 1050 |
*L = piv; |
| 1051 |
|
| 1052 |
++i; |
| 1053 |
rng [i].l = L + 1; |
| 1054 |
rng [i].r = rng [i - 1].r; |
| 1055 |
rng [i - 1].r = L; |
| 1056 |
|
| 1057 |
if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l) |
| 1058 |
{ |
| 1059 |
struct rng t; |
| 1060 |
|
| 1061 |
t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t; |
| 1062 |
} |
| 1063 |
} |
| 1064 |
else |
| 1065 |
--i; |
| 1066 |
} |
| 1067 |
} |
| 1068 |
|
| 1069 |
/* use an insertion sort after qsort, or for small arrays */ |
| 1070 |
/* first move the smallest element to the front, to act as a sentinel */ |
| 1071 |
{ |
| 1072 |
int i; |
| 1073 |
eio_dirent *min = dents; |
| 1074 |
|
| 1075 |
for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF + 1 : size; --i; ) |
| 1076 |
if (EIO_DENT_CMP (dents [i], <, *min)) |
| 1077 |
min = &dents [i]; |
| 1078 |
|
| 1079 |
/* swap elements 0 and j (minimum) */ |
| 1080 |
{ |
| 1081 |
eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
| 1082 |
} |
| 1083 |
} |
| 1084 |
|
| 1085 |
{ |
| 1086 |
/* then do standard insertion sort */ |
| 1087 |
eio_dirent *i, *j; |
| 1088 |
|
| 1089 |
for (i = dents + 1; i < dents + size; ++i) |
| 1090 |
{ |
| 1091 |
eio_dirent value = *i; |
| 1092 |
|
| 1093 |
for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
| 1094 |
j [1] = j [0]; |
| 1095 |
|
| 1096 |
j [1] = value; |
| 1097 |
} |
| 1098 |
} |
| 1099 |
} |
| 1100 |
|
| 1101 |
/* read a full directory */ |
| 1102 |
static void |
| 1103 |
eio__scandir (eio_req *req, etp_worker *self) |
| 1104 |
{ |
| 1105 |
DIR *dirp; |
| 1106 |
EIO_STRUCT_DIRENT *entp; |
| 1107 |
char *name, *names; |
| 1108 |
int namesalloc = 4096; |
| 1109 |
int namesoffs = 0; |
| 1110 |
int flags = req->int1; |
| 1111 |
eio_dirent *dents = 0; |
| 1112 |
int dentalloc = 128; |
| 1113 |
int dentoffs = 0; |
| 1114 |
|
| 1115 |
req->result = -1; |
| 1116 |
|
| 1117 |
if (!(flags & EIO_READDIR_DENTS)) |
| 1118 |
flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
| 1119 |
|
| 1120 |
X_LOCK (wrklock); |
| 1121 |
/* the corresponding closedir is in ETP_WORKER_CLEAR */ |
| 1122 |
self->dirp = dirp = opendir (req->ptr1); |
| 1123 |
req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
| 1124 |
req->ptr1 = names = malloc (namesalloc); |
| 1125 |
req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
| 1126 |
X_UNLOCK (wrklock); |
| 1127 |
|
| 1128 |
if (dirp && names && (!flags || dents)) |
| 1129 |
for (;;) |
| 1130 |
{ |
| 1131 |
errno = 0; |
| 1132 |
entp = readdir (dirp); |
| 1133 |
|
| 1134 |
if (!entp) |
| 1135 |
{ |
| 1136 |
if (errno) |
| 1137 |
break; |
| 1138 |
|
| 1139 |
/* sort etc. */ |
| 1140 |
req->int1 = flags; |
| 1141 |
req->result = dentoffs; |
| 1142 |
|
| 1143 |
if (dents) |
| 1144 |
{ |
| 1145 |
eio_dirent *ent = dents + dentoffs; |
| 1146 |
|
| 1147 |
while (ent > dents) |
| 1148 |
(--ent)->name = names + (size_t)ent->name; |
| 1149 |
} |
| 1150 |
|
| 1151 |
if (flags & EIO_READDIR_STAT_ORDER |
| 1152 |
|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
| 1153 |
eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */ |
| 1154 |
else if (flags & EIO_READDIR_DIRS_FIRST) |
| 1155 |
{ |
| 1156 |
/* in this case, all is known, and we just put dirs first and sort them */ |
| 1157 |
eio_dirent *ent = dents + dentoffs; |
| 1158 |
eio_dirent *dir = dents; |
| 1159 |
|
| 1160 |
/* now move dirs to the front, and non-dirs to the back */ |
| 1161 |
/* by walking from both sides and swapping if necessary */ |
| 1162 |
while (ent > dir) |
| 1163 |
{ |
| 1164 |
if (dir->type == DT_DIR) |
| 1165 |
{ |
| 1166 |
dir->score = 0; |
| 1167 |
++dir; |
| 1168 |
} |
| 1169 |
else |
| 1170 |
{ |
| 1171 |
--ent; |
| 1172 |
|
| 1173 |
if (ent->type == DT_DIR) |
| 1174 |
{ |
| 1175 |
eio_dirent tmp = *dir; |
| 1176 |
*dir = *ent; |
| 1177 |
*ent = tmp; |
| 1178 |
|
| 1179 |
dir->score = 0; |
| 1180 |
++dir; |
| 1181 |
} |
| 1182 |
} |
| 1183 |
} |
| 1184 |
|
| 1185 |
/* now sort the dirs only */ |
| 1186 |
eio_dent_sort (dents, dir - dents); |
| 1187 |
} |
| 1188 |
|
| 1189 |
/* only provide the names array unless DENTS is specified */ |
| 1190 |
if (!(flags & EIO_READDIR_DENTS)) |
| 1191 |
{ |
| 1192 |
X_LOCK (wrklock); |
| 1193 |
assert (!dents); |
| 1194 |
req->ptr1 = 0; |
| 1195 |
req->ptr2 = names; |
| 1196 |
X_UNLOCK (wrklock); |
| 1197 |
} |
| 1198 |
|
| 1199 |
break; |
| 1200 |
} |
| 1201 |
|
| 1202 |
/* now add the entry to our list(s) */ |
| 1203 |
name = entp->d_name; |
| 1204 |
|
| 1205 |
/* skip . and .. entries */ |
| 1206 |
if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
| 1207 |
{ |
| 1208 |
int len = D_NAMLEN (entp) + 1; |
| 1209 |
|
| 1210 |
while (expect_false (namesoffs + len > namesalloc)) |
| 1211 |
{ |
| 1212 |
namesalloc *= 2; |
| 1213 |
X_LOCK (wrklock); |
| 1214 |
req->ptr1 = names = realloc (names, namesalloc); |
| 1215 |
X_UNLOCK (wrklock); |
| 1216 |
|
| 1217 |
if (!names) |
| 1218 |
break; |
| 1219 |
} |
| 1220 |
|
| 1221 |
memcpy (names + namesoffs, name, len); |
| 1222 |
|
| 1223 |
if (dents) |
| 1224 |
{ |
| 1225 |
struct eio_dirent *ent; |
| 1226 |
|
| 1227 |
if (expect_false (dentoffs == dentalloc)) |
| 1228 |
{ |
| 1229 |
dentalloc *= 2; |
| 1230 |
X_LOCK (wrklock); |
| 1231 |
req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
| 1232 |
X_UNLOCK (wrklock); |
| 1233 |
|
| 1234 |
if (!dents) |
| 1235 |
break; |
| 1236 |
} |
| 1237 |
|
| 1238 |
ent = dents + dentoffs; |
| 1239 |
|
| 1240 |
ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
| 1241 |
ent->namelen = len - 1; |
| 1242 |
ent->inode = D_INO (entp) ? D_INO (entp) : dentoffs; |
| 1243 |
|
| 1244 |
switch (D_TYPE (entp)) |
| 1245 |
{ |
| 1246 |
default: |
| 1247 |
ent->type = EIO_DT_UNKNOWN; |
| 1248 |
flags |= EIO_READDIR_FOUND_UNKNOWN; |
| 1249 |
break; |
| 1250 |
|
| 1251 |
#ifdef DT_FIFO |
| 1252 |
case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
| 1253 |
#endif |
| 1254 |
#ifdef DT_CHR |
| 1255 |
case DT_CHR: ent->type = EIO_DT_CHR; break; |
| 1256 |
#endif |
| 1257 |
#ifdef DT_MPC |
| 1258 |
case DT_MPC: ent->type = EIO_DT_MPC; break; |
| 1259 |
#endif |
| 1260 |
#ifdef DT_DIR |
| 1261 |
case DT_DIR: ent->type = EIO_DT_DIR; break; |
| 1262 |
#endif |
| 1263 |
#ifdef DT_NAM |
| 1264 |
case DT_NAM: ent->type = EIO_DT_NAM; break; |
| 1265 |
#endif |
| 1266 |
#ifdef DT_BLK |
| 1267 |
case DT_BLK: ent->type = EIO_DT_BLK; break; |
| 1268 |
#endif |
| 1269 |
#ifdef DT_MPB |
| 1270 |
case DT_MPB: ent->type = EIO_DT_MPB; break; |
| 1271 |
#endif |
| 1272 |
#ifdef DT_REG |
| 1273 |
case DT_REG: ent->type = EIO_DT_REG; break; |
| 1274 |
#endif |
| 1275 |
#ifdef DT_NWK |
| 1276 |
case DT_NWK: ent->type = EIO_DT_NWK; break; |
| 1277 |
#endif |
| 1278 |
#ifdef DT_CMP |
| 1279 |
case DT_CMP: ent->type = EIO_DT_CMP; break; |
| 1280 |
#endif |
| 1281 |
#ifdef DT_LNK |
| 1282 |
case DT_LNK: ent->type = EIO_DT_LNK; break; |
| 1283 |
#endif |
| 1284 |
#ifdef DT_SOCK |
| 1285 |
case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
| 1286 |
#endif |
| 1287 |
#ifdef DT_DOOR |
| 1288 |
case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
| 1289 |
#endif |
| 1290 |
#ifdef DT_WHT |
| 1291 |
case DT_WHT: ent->type = EIO_DT_WHT; break; |
| 1292 |
#endif |
| 1293 |
} |
| 1294 |
|
| 1295 |
ent->score = 0; |
| 1296 |
|
| 1297 |
if (flags & EIO_READDIR_DIRS_FIRST) |
| 1298 |
{ |
| 1299 |
if (ent->type == EIO_DT_UNKNOWN) |
| 1300 |
{ |
| 1301 |
if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
| 1302 |
ent->score = 7; |
| 1303 |
else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
| 1304 |
ent->score = len <= 2 ? len + 4 : len <= 4 ? 3 : len <= 7 ? 2 : 1; /* shorter == more likely dir, but avoid too many classes */ |
| 1305 |
} |
| 1306 |
else if (ent->type == EIO_DT_DIR) |
| 1307 |
ent->score = 8; |
| 1308 |
} |
| 1309 |
} |
| 1310 |
|
| 1311 |
namesoffs += len; |
| 1312 |
++dentoffs; |
| 1313 |
} |
| 1314 |
|
| 1315 |
if (EIO_CANCELLED (req)) |
| 1316 |
{ |
| 1317 |
errno = ECANCELED; |
| 1318 |
break; |
| 1319 |
} |
| 1320 |
} |
| 1321 |
} |
| 1322 |
|
| 1323 |
#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
| 1324 |
# undef msync |
| 1325 |
# define msync(a,b,c) ((errno = ENOSYS), -1) |
| 1326 |
#endif |
| 1327 |
|
| 1328 |
int |
| 1329 |
eio__mtouch (void *mem, size_t len, int flags) |
| 1330 |
{ |
| 1331 |
intptr_t addr = (intptr_t)mem; |
| 1332 |
intptr_t end = addr + len; |
| 1333 |
#ifdef PAGESIZE |
| 1334 |
const intptr_t page = PAGESIZE; |
| 1335 |
#else |
| 1336 |
static intptr_t page; |
| 1337 |
|
| 1338 |
if (!page) |
| 1339 |
page = sysconf (_SC_PAGESIZE); |
| 1340 |
#endif |
| 1341 |
|
| 1342 |
addr &= ~(page - 1); /* assume page size is always a power of two */ |
| 1343 |
|
| 1344 |
if (addr < end) |
| 1345 |
if (flags) /* modify */ |
| 1346 |
do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
| 1347 |
else |
| 1348 |
do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
| 1349 |
|
| 1350 |
return 0; |
| 1351 |
} |
| 1352 |
|
| 1353 |
/*****************************************************************************/ |
| 1354 |
|
| 1355 |
#define ALLOC(len) \ |
| 1356 |
if (!req->ptr2) \ |
| 1357 |
{ \ |
| 1358 |
X_LOCK (wrklock); \ |
| 1359 |
req->flags |= EIO_FLAG_PTR2_FREE; \ |
| 1360 |
X_UNLOCK (wrklock); \ |
| 1361 |
req->ptr2 = malloc (len); \ |
| 1362 |
if (!req->ptr2) \ |
| 1363 |
{ \ |
| 1364 |
errno = ENOMEM; \ |
| 1365 |
req->result = -1; \ |
| 1366 |
break; \ |
| 1367 |
} \ |
| 1368 |
} |
| 1369 |
|
| 1370 |
X_THREAD_PROC (etp_proc) |
| 1371 |
{ |
| 1372 |
ETP_REQ *req; |
| 1373 |
struct timespec ts; |
| 1374 |
etp_worker *self = (etp_worker *)thr_arg; |
| 1375 |
|
| 1376 |
/* try to distribute timeouts somewhat randomly */ |
| 1377 |
ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
| 1378 |
|
| 1379 |
for (;;) |
| 1380 |
{ |
| 1381 |
X_LOCK (reqlock); |
| 1382 |
|
| 1383 |
for (;;) |
| 1384 |
{ |
| 1385 |
self->req = req = reqq_shift (&req_queue); |
| 1386 |
|
| 1387 |
if (req) |
| 1388 |
break; |
| 1389 |
|
| 1390 |
++idle; |
| 1391 |
|
| 1392 |
ts.tv_sec = time (0) + IDLE_TIMEOUT; |
| 1393 |
if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
| 1394 |
{ |
| 1395 |
if (idle > max_idle) |
| 1396 |
{ |
| 1397 |
--idle; |
| 1398 |
X_UNLOCK (reqlock); |
| 1399 |
X_LOCK (wrklock); |
| 1400 |
--started; |
| 1401 |
X_UNLOCK (wrklock); |
| 1402 |
goto quit; |
| 1403 |
} |
| 1404 |
|
| 1405 |
/* we are allowed to idle, so do so without any timeout */ |
| 1406 |
X_COND_WAIT (reqwait, reqlock); |
| 1407 |
} |
| 1408 |
|
| 1409 |
--idle; |
| 1410 |
} |
| 1411 |
|
| 1412 |
--nready; |
| 1413 |
|
| 1414 |
X_UNLOCK (reqlock); |
| 1415 |
|
| 1416 |
if (req->type < 0) |
| 1417 |
goto quit; |
| 1418 |
|
| 1419 |
if (!EIO_CANCELLED (req)) |
| 1420 |
ETP_EXECUTE (self, req); |
| 1421 |
|
| 1422 |
X_LOCK (reslock); |
| 1423 |
|
| 1424 |
++npending; |
| 1425 |
|
| 1426 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
| 1427 |
want_poll_cb (); |
| 1428 |
|
| 1429 |
self->req = 0; |
| 1430 |
etp_worker_clear (self); |
| 1431 |
|
| 1432 |
X_UNLOCK (reslock); |
| 1433 |
} |
| 1434 |
|
| 1435 |
quit: |
| 1436 |
X_LOCK (wrklock); |
| 1437 |
etp_worker_free (self); |
| 1438 |
X_UNLOCK (wrklock); |
| 1439 |
|
| 1440 |
return 0; |
| 1441 |
} |
| 1442 |
|
| 1443 |
/*****************************************************************************/ |
| 1444 |
|
| 1445 |
int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
| 1446 |
{ |
| 1447 |
return etp_init (want_poll, done_poll); |
| 1448 |
} |
| 1449 |
|
| 1450 |
static void eio_api_destroy (eio_req *req) |
| 1451 |
{ |
| 1452 |
free (req); |
| 1453 |
} |
| 1454 |
|
| 1455 |
#define REQ(rtype) \ |
| 1456 |
eio_req *req; \ |
| 1457 |
\ |
| 1458 |
req = (eio_req *)calloc (1, sizeof *req); \ |
| 1459 |
if (!req) \ |
| 1460 |
return 0; \ |
| 1461 |
\ |
| 1462 |
req->type = rtype; \ |
| 1463 |
req->pri = pri; \ |
| 1464 |
req->finish = cb; \ |
| 1465 |
req->data = data; \ |
| 1466 |
req->destroy = eio_api_destroy; |
| 1467 |
|
| 1468 |
#define SEND eio_submit (req); return req |
| 1469 |
|
| 1470 |
#define PATH \ |
| 1471 |
req->flags |= EIO_FLAG_PTR1_FREE; \ |
| 1472 |
req->ptr1 = strdup (path); \ |
| 1473 |
if (!req->ptr1) \ |
| 1474 |
{ \ |
| 1475 |
eio_api_destroy (req); \ |
| 1476 |
return 0; \ |
| 1477 |
} |
| 1478 |
|
| 1479 |
static void eio_execute (etp_worker *self, eio_req *req) |
| 1480 |
{ |
| 1481 |
errno = 0; |
| 1482 |
|
| 1483 |
switch (req->type) |
| 1484 |
{ |
| 1485 |
case EIO_READ: ALLOC (req->size); |
| 1486 |
req->result = req->offs >= 0 |
| 1487 |
? pread (req->int1, req->ptr2, req->size, req->offs) |
| 1488 |
: read (req->int1, req->ptr2, req->size); break; |
| 1489 |
case EIO_WRITE: req->result = req->offs >= 0 |
| 1490 |
? pwrite (req->int1, req->ptr2, req->size, req->offs) |
| 1491 |
: write (req->int1, req->ptr2, req->size); break; |
| 1492 |
|
| 1493 |
case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
| 1494 |
case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
| 1495 |
|
| 1496 |
case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
| 1497 |
req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
| 1498 |
case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
| 1499 |
req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
| 1500 |
case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
| 1501 |
req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
| 1502 |
|
| 1503 |
case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
| 1504 |
case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
| 1505 |
case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
| 1506 |
case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
| 1507 |
case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
| 1508 |
case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
| 1509 |
|
| 1510 |
case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
| 1511 |
case EIO_CLOSE: req->result = close (req->int1); break; |
| 1512 |
case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
| 1513 |
case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
| 1514 |
case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
| 1515 |
case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
| 1516 |
case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
| 1517 |
case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
| 1518 |
case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
| 1519 |
case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
| 1520 |
|
| 1521 |
case EIO_READLINK: ALLOC (NAME_MAX); |
| 1522 |
req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
| 1523 |
|
| 1524 |
case EIO_SYNC: req->result = 0; sync (); break; |
| 1525 |
case EIO_FSYNC: req->result = fsync (req->int1); break; |
| 1526 |
case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
| 1527 |
case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
| 1528 |
case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
| 1529 |
case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
| 1530 |
|
| 1531 |
case EIO_READDIR: eio__scandir (req, self); break; |
| 1532 |
|
| 1533 |
case EIO_BUSY: |
| 1534 |
#ifdef _WIN32 |
| 1535 |
Sleep (req->nv1 * 1000.); |
| 1536 |
#else |
| 1537 |
{ |
| 1538 |
struct timeval tv; |
| 1539 |
|
| 1540 |
tv.tv_sec = req->nv1; |
| 1541 |
tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
| 1542 |
|
| 1543 |
req->result = select (0, 0, 0, 0, &tv); |
| 1544 |
} |
| 1545 |
#endif |
| 1546 |
break; |
| 1547 |
|
| 1548 |
case EIO_UTIME: |
| 1549 |
case EIO_FUTIME: |
| 1550 |
{ |
| 1551 |
struct timeval tv[2]; |
| 1552 |
struct timeval *times; |
| 1553 |
|
| 1554 |
if (req->nv1 != -1. || req->nv2 != -1.) |
| 1555 |
{ |
| 1556 |
tv[0].tv_sec = req->nv1; |
| 1557 |
tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
| 1558 |
tv[1].tv_sec = req->nv2; |
| 1559 |
tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
| 1560 |
|
| 1561 |
times = tv; |
| 1562 |
} |
| 1563 |
else |
| 1564 |
times = 0; |
| 1565 |
|
| 1566 |
|
| 1567 |
req->result = req->type == EIO_FUTIME |
| 1568 |
? futimes (req->int1, times) |
| 1569 |
: utimes (req->ptr1, times); |
| 1570 |
} |
| 1571 |
break; |
| 1572 |
|
| 1573 |
case EIO_GROUP: |
| 1574 |
abort (); /* handled in eio_request */ |
| 1575 |
|
| 1576 |
case EIO_NOP: |
| 1577 |
req->result = 0; |
| 1578 |
break; |
| 1579 |
|
| 1580 |
case EIO_CUSTOM: |
| 1581 |
((void (*)(eio_req *))req->feed) (req); |
| 1582 |
break; |
| 1583 |
|
| 1584 |
default: |
| 1585 |
req->result = -1; |
| 1586 |
break; |
| 1587 |
} |
| 1588 |
|
| 1589 |
req->errorno = errno; |
| 1590 |
} |
| 1591 |
|
| 1592 |
#ifndef EIO_NO_WRAPPERS |
| 1593 |
|
| 1594 |
eio_req *eio_nop (int pri, eio_cb cb, void *data) |
| 1595 |
{ |
| 1596 |
REQ (EIO_NOP); SEND; |
| 1597 |
} |
| 1598 |
|
| 1599 |
eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data) |
| 1600 |
{ |
| 1601 |
REQ (EIO_BUSY); req->nv1 = delay; SEND; |
| 1602 |
} |
| 1603 |
|
| 1604 |
eio_req *eio_sync (int pri, eio_cb cb, void *data) |
| 1605 |
{ |
| 1606 |
REQ (EIO_SYNC); SEND; |
| 1607 |
} |
| 1608 |
|
| 1609 |
eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
| 1610 |
{ |
| 1611 |
REQ (EIO_FSYNC); req->int1 = fd; SEND; |
| 1612 |
} |
| 1613 |
|
| 1614 |
eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
| 1615 |
{ |
| 1616 |
REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
| 1617 |
} |
| 1618 |
|
| 1619 |
eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
| 1620 |
{ |
| 1621 |
REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
| 1622 |
} |
| 1623 |
|
| 1624 |
eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
| 1625 |
{ |
| 1626 |
REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
| 1627 |
} |
| 1628 |
|
| 1629 |
eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
| 1630 |
{ |
| 1631 |
REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
| 1632 |
} |
| 1633 |
|
| 1634 |
eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
| 1635 |
{ |
| 1636 |
REQ (EIO_CLOSE); req->int1 = fd; SEND; |
| 1637 |
} |
| 1638 |
|
| 1639 |
eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) |
| 1640 |
{ |
| 1641 |
REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; |
| 1642 |
} |
| 1643 |
|
| 1644 |
eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
| 1645 |
{ |
| 1646 |
REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
| 1647 |
} |
| 1648 |
|
| 1649 |
eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
| 1650 |
{ |
| 1651 |
REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
| 1652 |
} |
| 1653 |
|
| 1654 |
eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
| 1655 |
{ |
| 1656 |
REQ (EIO_FSTAT); req->int1 = fd; SEND; |
| 1657 |
} |
| 1658 |
|
| 1659 |
eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
| 1660 |
{ |
| 1661 |
REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
| 1662 |
} |
| 1663 |
|
| 1664 |
eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data) |
| 1665 |
{ |
| 1666 |
REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; |
| 1667 |
} |
| 1668 |
|
| 1669 |
eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
| 1670 |
{ |
| 1671 |
REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
| 1672 |
} |
| 1673 |
|
| 1674 |
eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
| 1675 |
{ |
| 1676 |
REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
| 1677 |
} |
| 1678 |
|
| 1679 |
eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
| 1680 |
{ |
| 1681 |
REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; |
| 1682 |
} |
| 1683 |
|
| 1684 |
eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data) |
| 1685 |
{ |
| 1686 |
REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; |
| 1687 |
} |
| 1688 |
|
| 1689 |
eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data) |
| 1690 |
{ |
| 1691 |
REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; |
| 1692 |
} |
| 1693 |
|
| 1694 |
eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data) |
| 1695 |
{ |
| 1696 |
REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; |
| 1697 |
} |
| 1698 |
|
| 1699 |
eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
| 1700 |
{ |
| 1701 |
REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
| 1702 |
} |
| 1703 |
|
| 1704 |
eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
| 1705 |
{ |
| 1706 |
REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
| 1707 |
} |
| 1708 |
|
| 1709 |
eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
| 1710 |
{ |
| 1711 |
REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; |
| 1712 |
} |
| 1713 |
|
| 1714 |
eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
| 1715 |
{ |
| 1716 |
REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; |
| 1717 |
} |
| 1718 |
|
| 1719 |
static eio_req * |
| 1720 |
eio__1path (int type, const char *path, int pri, eio_cb cb, void *data) |
| 1721 |
{ |
| 1722 |
REQ (type); PATH; SEND; |
| 1723 |
} |
| 1724 |
|
| 1725 |
eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
| 1726 |
{ |
| 1727 |
return eio__1path (EIO_READLINK, path, pri, cb, data); |
| 1728 |
} |
| 1729 |
|
| 1730 |
eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
| 1731 |
{ |
| 1732 |
return eio__1path (EIO_STAT, path, pri, cb, data); |
| 1733 |
} |
| 1734 |
|
| 1735 |
eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
| 1736 |
{ |
| 1737 |
return eio__1path (EIO_LSTAT, path, pri, cb, data); |
| 1738 |
} |
| 1739 |
|
| 1740 |
eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
| 1741 |
{ |
| 1742 |
return eio__1path (EIO_UNLINK, path, pri, cb, data); |
| 1743 |
} |
| 1744 |
|
| 1745 |
eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
| 1746 |
{ |
| 1747 |
return eio__1path (EIO_RMDIR, path, pri, cb, data); |
| 1748 |
} |
| 1749 |
|
| 1750 |
eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
| 1751 |
{ |
| 1752 |
REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
| 1753 |
} |
| 1754 |
|
| 1755 |
eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
| 1756 |
{ |
| 1757 |
REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
| 1758 |
} |
| 1759 |
|
| 1760 |
static eio_req * |
| 1761 |
eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
| 1762 |
{ |
| 1763 |
REQ (type); PATH; |
| 1764 |
|
| 1765 |
req->flags |= EIO_FLAG_PTR2_FREE; |
| 1766 |
req->ptr2 = strdup (new_path); |
| 1767 |
if (!req->ptr2) |
| 1768 |
{ |
| 1769 |
eio_api_destroy (req); |
| 1770 |
return 0; |
| 1771 |
} |
| 1772 |
|
| 1773 |
SEND; |
| 1774 |
} |
| 1775 |
|
| 1776 |
eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
| 1777 |
{ |
| 1778 |
return eio__2path (EIO_LINK, path, new_path, pri, cb, data); |
| 1779 |
} |
| 1780 |
|
| 1781 |
eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
| 1782 |
{ |
| 1783 |
return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data); |
| 1784 |
} |
| 1785 |
|
| 1786 |
eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
| 1787 |
{ |
| 1788 |
return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
| 1789 |
} |
| 1790 |
|
| 1791 |
eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
| 1792 |
{ |
| 1793 |
REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
| 1794 |
} |
| 1795 |
|
| 1796 |
#endif |
| 1797 |
|
| 1798 |
eio_req *eio_grp (eio_cb cb, void *data) |
| 1799 |
{ |
| 1800 |
const int pri = EIO_PRI_MAX; |
| 1801 |
|
| 1802 |
REQ (EIO_GROUP); SEND; |
| 1803 |
} |
| 1804 |
|
| 1805 |
#undef REQ |
| 1806 |
#undef PATH |
| 1807 |
#undef SEND |
| 1808 |
|
| 1809 |
/*****************************************************************************/ |
| 1810 |
/* grp functions */ |
| 1811 |
|
| 1812 |
void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
| 1813 |
{ |
| 1814 |
grp->int2 = limit; |
| 1815 |
grp->feed = feed; |
| 1816 |
|
| 1817 |
grp_try_feed (grp); |
| 1818 |
} |
| 1819 |
|
| 1820 |
void eio_grp_limit (eio_req *grp, int limit) |
| 1821 |
{ |
| 1822 |
grp->int2 = limit; |
| 1823 |
|
| 1824 |
grp_try_feed (grp); |
| 1825 |
} |
| 1826 |
|
| 1827 |
void eio_grp_add (eio_req *grp, eio_req *req) |
| 1828 |
{ |
| 1829 |
assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
| 1830 |
|
| 1831 |
grp->flags |= EIO_FLAG_GROUPADD; |
| 1832 |
|
| 1833 |
++grp->size; |
| 1834 |
req->grp = grp; |
| 1835 |
|
| 1836 |
req->grp_prev = 0; |
| 1837 |
req->grp_next = grp->grp_first; |
| 1838 |
|
| 1839 |
if (grp->grp_first) |
| 1840 |
grp->grp_first->grp_prev = req; |
| 1841 |
|
| 1842 |
grp->grp_first = req; |
| 1843 |
} |
| 1844 |
|
| 1845 |
/*****************************************************************************/ |
| 1846 |
/* misc garbage */ |
| 1847 |
|
| 1848 |
ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
| 1849 |
{ |
| 1850 |
etp_worker wrk; |
| 1851 |
ssize_t ret; |
| 1852 |
|
| 1853 |
wrk.dbuf = 0; |
| 1854 |
|
| 1855 |
ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
| 1856 |
|
| 1857 |
if (wrk.dbuf) |
| 1858 |
free (wrk.dbuf); |
| 1859 |
|
| 1860 |
return ret; |
| 1861 |
} |
| 1862 |
|
