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

Comparing libeio/eio.c (file contents):
Revision 1.91 by root, Sun Jul 17 04:20:04 2011 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 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- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 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- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
49#endif 49#endif
50#include "xthread.h" 50#include "xthread.h"
51 51
52#include <errno.h> 52#include <errno.h>
53#include <stddef.h> 53#include <stddef.h>
71#endif 71#endif
72#ifndef ELOOP 72#ifndef ELOOP
73# define ELOOP EDOM 73# define ELOOP EDOM
74#endif 74#endif
75 75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
76static void eio_destroy (eio_req *req); 80static void eio_destroy (eio_req *req);
77 81
78#ifndef EIO_FINISH 82#ifndef EIO_FINISH
79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
80#endif 84#endif
101#ifdef _WIN32 105#ifdef _WIN32
102 106
103 #undef PAGESIZE 107 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */ 108 #define PAGESIZE 4096 /* GetSystemInfo? */
105 109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
106 #ifdef EIO_STRUCT_STATI64 112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
107 #define stat(path,buf) _stati64 (path,buf) 114 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf) 115 #define fstat(fd,buf) _fstati64 (fd,buf)
109 #endif 116 #endif
110 #define lstat(path,buf) stat (path,buf) 117 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path) 119 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114 121
115 #define chmod(path,mode) _chmod (path, mode) 122 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd) 123 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2) 124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
118 126
119 #define fchmod(fd,mode) EIO_ENOSYS () 127 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS () 128 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS () 129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
125 #define sync() EIO_ENOSYS () 133 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS () 134 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS () 135 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS () 136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129 137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
130 /* we could even stat and see if it exists */ 204 /* we could even stat and see if it exists */
131 static int 205 static int
132 symlink (const char *old, const char *neu) 206 symlink (const char *old, const char *neu)
133 { 207 {
134 #if WINVER >= 0x0600 208 #if WINVER >= 0x0600
140 #endif 214 #endif
141 215
142 return EIO_ERRNO (ENOENT, -1); 216 return EIO_ERRNO (ENOENT, -1);
143 } 217 }
144 218
145 /* POSIX API only */ 219 /* POSIX API only, causing trouble for win32 apps */
146 #define CreateHardLink(neu,old,flags) 0 220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
147 #define CreateSymbolicLink(neu,old,flags) 0 221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
148 222
149 struct statvfs 223 struct statvfs
150 { 224 {
151 int dummy; 225 int dummy;
152 }; 226 };
158 232
159#else 233#else
160 234
161 #include <sys/time.h> 235 #include <sys/time.h>
162 #include <sys/select.h> 236 #include <sys/select.h>
163 #include <sys/statvfs.h>
164 #include <unistd.h> 237 #include <unistd.h>
165 #include <utime.h>
166 #include <signal.h> 238 #include <signal.h>
167 #include <dirent.h> 239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
168 249
169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
170 #include <sys/mman.h> 251 #include <sys/mman.h>
171 #endif 252 #endif
172 253
173 #define D_NAME(entp) entp->d_name 254 #define D_NAME(entp) entp->d_name
174 255
175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
177 #define _DIRENT_HAVE_D_TYPE /* sigh */ 258 #define _DIRENT_HAVE_D_TYPE /* sigh */
178 #define D_INO(de) (de)->d_fileno 259 #define D_INO(de) (de)->d_fileno
179 #define D_NAMLEN(de) (de)->d_namlen 260 #define D_NAMLEN(de) (de)->d_namlen
180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
181 #define D_INO(de) (de)->d_ino 262 #define D_INO(de) (de)->d_ino
192 273
193 #ifndef EIO_STRUCT_DIRENT 274 #ifndef EIO_STRUCT_DIRENT
194 #define EIO_STRUCT_DIRENT struct dirent 275 #define EIO_STRUCT_DIRENT struct dirent
195 #endif 276 #endif
196 277
278#endif
279
280#if HAVE_UTIMES
281# include <utime.h>
282#endif
283
284#if HAVE_SYS_SYSCALL_H
285# include <sys/syscall.h>
286#endif
287
288#if HAVE_SYS_PRCTL_H
289# include <sys/prctl.h>
197#endif 290#endif
198 291
199#if HAVE_SENDFILE 292#if HAVE_SENDFILE
200# if __linux 293# if __linux
201# include <sys/sendfile.h> 294# include <sys/sendfile.h>
232#endif 325#endif
233 326
234/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
235#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
236 329
237#define dBUF \ 330#define dBUF \
238 char *eio_buf; \
239 ETP_WORKER_LOCK (self); \
240 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
241 ETP_WORKER_UNLOCK (self); \
242 errno = ENOMEM; \ 332 errno = ENOMEM; \
243 if (!eio_buf) \ 333 if (!eio_buf) \
244 return -1; 334 return -1
245 335
246#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
338
339/*****************************************************************************/
340
341struct etp_tmpbuf;
342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
349#else
350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
352#endif
353
354struct eio_pwd
355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
362
363/*****************************************************************************/
247 364
248#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
249#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
250 367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
370
371static void eio_nop_callback (void) { }
372static void (*eio_want_poll_cb)(void) = eio_nop_callback;
373static void (*eio_done_poll_cb)(void) = eio_nop_callback;
374
375#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
376#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
377
251struct etp_worker; 378struct etp_worker;
252
253#define ETP_REQ eio_req 379#define ETP_REQ eio_req
254#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
255static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
256#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
257static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
258#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
259 385
260#define ETP_WORKER_CLEAR(req) \ 386#include "etp.c"
261 if (wrk->dbuf) \
262 { \
263 free (wrk->dbuf); \
264 wrk->dbuf = 0; \
265 }
266 387
267#define ETP_WORKER_COMMON \ 388static struct etp_pool eio_pool;
268 void *dbuf; 389#define EIO_POOL (&eio_pool)
269
270/*****************************************************************************/
271
272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
273
274/* calculate time difference in ~1/EIO_TICKS of a second */
275ecb_inline int
276tvdiff (struct timeval *tv1, struct timeval *tv2)
277{
278 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
279 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
280}
281
282static unsigned int started, idle, wanted = 4;
283
284static void (*want_poll_cb) (void);
285static void (*done_poll_cb) (void);
286
287static unsigned int max_poll_time; /* reslock */
288static unsigned int max_poll_reqs; /* reslock */
289
290static volatile unsigned int nreqs; /* reqlock */
291static volatile unsigned int nready; /* reqlock */
292static volatile unsigned int npending; /* reqlock */
293static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
294static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
295
296static xmutex_t wrklock;
297static xmutex_t reslock;
298static xmutex_t reqlock;
299static xcond_t reqwait;
300
301#if !HAVE_PREADWRITE
302/*
303 * make our pread/pwrite emulation safe against themselves, but not against
304 * normal read/write by using a mutex. slows down execution a lot,
305 * but that's your problem, not mine.
306 */
307static xmutex_t preadwritelock = X_MUTEX_INIT;
308#endif
309
310typedef struct etp_worker
311{
312 /* locked by wrklock */
313 struct etp_worker *prev, *next;
314
315 xthread_t tid;
316
317 /* locked by reslock, reqlock or wrklock */
318 ETP_REQ *req; /* currently processed request */
319
320 ETP_WORKER_COMMON
321} etp_worker;
322
323static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
324
325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
327
328/* worker threads management */
329
330static void ecb_cold
331etp_worker_clear (etp_worker *wrk)
332{
333 ETP_WORKER_CLEAR (wrk);
334}
335
336static void ecb_cold
337etp_worker_free (etp_worker *wrk)
338{
339 wrk->next->prev = wrk->prev;
340 wrk->prev->next = wrk->next;
341
342 free (wrk);
343}
344
345static unsigned int
346etp_nreqs (void)
347{
348 int retval;
349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
350 retval = nreqs;
351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
352 return retval;
353}
354
355static unsigned int
356etp_nready (void)
357{
358 unsigned int retval;
359
360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
361 retval = nready;
362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
363
364 return retval;
365}
366
367static unsigned int
368etp_npending (void)
369{
370 unsigned int retval;
371
372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
373 retval = npending;
374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
375
376 return retval;
377}
378
379static unsigned int
380etp_nthreads (void)
381{
382 unsigned int retval;
383
384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
385 retval = started;
386 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
387
388 return retval;
389}
390
391/*
392 * a somewhat faster data structure might be nice, but
393 * with 8 priorities this actually needs <20 insns
394 * per shift, the most expensive operation.
395 */
396typedef struct {
397 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
398 int size;
399} etp_reqq;
400
401static etp_reqq req_queue;
402static etp_reqq res_queue;
403
404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
415static int ecb_noinline
416reqq_push (etp_reqq *q, ETP_REQ *req)
417{
418 int pri = req->pri;
419 req->next = 0;
420
421 if (q->qe[pri])
422 {
423 q->qe[pri]->next = req;
424 q->qe[pri] = req;
425 }
426 else
427 q->qe[pri] = q->qs[pri] = req;
428
429 return q->size++;
430}
431
432static ETP_REQ * ecb_noinline
433reqq_shift (etp_reqq *q)
434{
435 int pri;
436
437 if (!q->size)
438 return 0;
439
440 --q->size;
441
442 for (pri = ETP_NUM_PRI; pri--; )
443 {
444 eio_req *req = q->qs[pri];
445
446 if (req)
447 {
448 if (!(q->qs[pri] = (eio_req *)req->next))
449 q->qe[pri] = 0;
450
451 return req;
452 }
453 }
454
455 abort ();
456}
457
458static void ecb_cold
459etp_thread_init (void)
460{
461#if !HAVE_PREADWRITE
462 X_MUTEX_CREATE (preadwritelock);
463#endif
464 X_MUTEX_CREATE (wrklock);
465 X_MUTEX_CREATE (reslock);
466 X_MUTEX_CREATE (reqlock);
467 X_COND_CREATE (reqwait);
468}
469
470static void ecb_cold
471etp_atfork_child (void)
472{
473 reqq_init (&req_queue);
474 reqq_init (&res_queue);
475
476 wrk_first.next =
477 wrk_first.prev = &wrk_first;
478
479 started = 0;
480 idle = 0;
481 nreqs = 0;
482 nready = 0;
483 npending = 0;
484
485 etp_thread_init ();
486}
487
488static void ecb_cold
489etp_once_init (void)
490{
491 etp_thread_init ();
492 X_THREAD_ATFORK (0, 0, etp_atfork_child);
493}
494
495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
497{
498 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
499
500 pthread_once (&doinit, etp_once_init);
501
502 want_poll_cb = want_poll;
503 done_poll_cb = done_poll;
504
505 return 0;
506}
507
508X_THREAD_PROC (etp_proc);
509
510static void ecb_cold
511etp_start_thread (void)
512{
513 etp_worker *wrk = calloc (1, sizeof (etp_worker));
514
515 /*TODO*/
516 assert (("unable to allocate worker thread data", wrk));
517
518 X_LOCK (wrklock);
519
520 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
521 {
522 wrk->prev = &wrk_first;
523 wrk->next = wrk_first.next;
524 wrk_first.next->prev = wrk;
525 wrk_first.next = wrk;
526 ++started;
527 }
528 else
529 free (wrk);
530
531 X_UNLOCK (wrklock);
532}
533
534static void
535etp_maybe_start_thread (void)
536{
537 if (ecb_expect_true (etp_nthreads () >= wanted))
538 return;
539
540 /* todo: maybe use idle here, but might be less exact */
541 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
542 return;
543
544 etp_start_thread ();
545}
546
547static void ecb_cold
548etp_end_thread (void)
549{
550 eio_req *req = calloc (1, sizeof (eio_req));
551
552 req->type = -1;
553 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
554
555 X_LOCK (reqlock);
556 reqq_push (&req_queue, req);
557 X_COND_SIGNAL (reqwait);
558 X_UNLOCK (reqlock);
559
560 X_LOCK (wrklock);
561 --started;
562 X_UNLOCK (wrklock);
563}
564
565static int
566etp_poll (void)
567{
568 unsigned int maxreqs;
569 unsigned int maxtime;
570 struct timeval tv_start, tv_now;
571
572 X_LOCK (reslock);
573 maxreqs = max_poll_reqs;
574 maxtime = max_poll_time;
575 X_UNLOCK (reslock);
576
577 if (maxtime)
578 gettimeofday (&tv_start, 0);
579
580 for (;;)
581 {
582 ETP_REQ *req;
583
584 etp_maybe_start_thread ();
585
586 X_LOCK (reslock);
587 req = reqq_shift (&res_queue);
588
589 if (req)
590 {
591 --npending;
592
593 if (!res_queue.size && done_poll_cb)
594 done_poll_cb ();
595 }
596
597 X_UNLOCK (reslock);
598
599 if (!req)
600 return 0;
601
602 X_LOCK (reqlock);
603 --nreqs;
604 X_UNLOCK (reqlock);
605
606 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
607 {
608 req->int1 = 1; /* mark request as delayed */
609 continue;
610 }
611 else
612 {
613 int res = ETP_FINISH (req);
614 if (ecb_expect_false (res))
615 return res;
616 }
617
618 if (ecb_expect_false (maxreqs && !--maxreqs))
619 break;
620
621 if (maxtime)
622 {
623 gettimeofday (&tv_now, 0);
624
625 if (tvdiff (&tv_start, &tv_now) >= maxtime)
626 break;
627 }
628 }
629
630 errno = EAGAIN;
631 return -1;
632}
633
634static void
635etp_cancel (ETP_REQ *req)
636{
637 req->cancelled = 1;
638
639 eio_grp_cancel (req);
640}
641
642static void
643etp_submit (ETP_REQ *req)
644{
645 req->pri -= ETP_PRI_MIN;
646
647 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
648 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
649
650 if (ecb_expect_false (req->type == EIO_GROUP))
651 {
652 /* I hope this is worth it :/ */
653 X_LOCK (reqlock);
654 ++nreqs;
655 X_UNLOCK (reqlock);
656
657 X_LOCK (reslock);
658
659 ++npending;
660
661 if (!reqq_push (&res_queue, req) && want_poll_cb)
662 want_poll_cb ();
663
664 X_UNLOCK (reslock);
665 }
666 else
667 {
668 X_LOCK (reqlock);
669 ++nreqs;
670 ++nready;
671 reqq_push (&req_queue, req);
672 X_COND_SIGNAL (reqwait);
673 X_UNLOCK (reqlock);
674
675 etp_maybe_start_thread ();
676 }
677}
678
679static void ecb_cold
680etp_set_max_poll_time (double nseconds)
681{
682 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
683 max_poll_time = nseconds * EIO_TICKS;
684 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
685}
686
687static void ecb_cold
688etp_set_max_poll_reqs (unsigned int maxreqs)
689{
690 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
691 max_poll_reqs = maxreqs;
692 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
693}
694
695static void ecb_cold
696etp_set_max_idle (unsigned int nthreads)
697{
698 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
699 max_idle = nthreads;
700 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
701}
702
703static void ecb_cold
704etp_set_idle_timeout (unsigned int seconds)
705{
706 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
707 idle_timeout = seconds;
708 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
709}
710
711static void ecb_cold
712etp_set_min_parallel (unsigned int nthreads)
713{
714 if (wanted < nthreads)
715 wanted = nthreads;
716}
717
718static void ecb_cold
719etp_set_max_parallel (unsigned int nthreads)
720{
721 if (wanted > nthreads)
722 wanted = nthreads;
723
724 while (started > wanted)
725 etp_end_thread ();
726}
727 390
728/*****************************************************************************/ 391/*****************************************************************************/
729 392
730static void 393static void
731grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
732{ 395{
733 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
734 { 397 {
735 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
736 399
737 EIO_FEED (grp); 400 EIO_FEED (grp);
738 401
739 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
740 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
741 { 404 {
742 grp->feed = 0; 405 grp->feed = 0;
743 break; 406 break;
744 } 407 }
745 } 408 }
752 415
753 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
754 grp_try_feed (grp); 417 grp_try_feed (grp);
755 418
756 /* finish, if done */ 419 /* finish, if done */
757 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
758 return eio_finish (grp); 421 return eio_finish (grp);
759 else 422 else
760 return 0; 423 return 0;
761} 424}
762 425
798} 461}
799 462
800void 463void
801eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
802{ 465{
803 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
804 eio_cancel (grp);
805} 467}
806 468
807void 469void
808eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
809{ 471{
810 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
811} 473}
812 474
813void 475void
814eio_submit (eio_req *req) 476eio_submit (eio_req *req)
815{ 477{
816 etp_submit (req); 478 etp_submit (EIO_POOL, req);
817} 479}
818 480
819unsigned int 481unsigned int
820eio_nreqs (void) 482eio_nreqs (void)
821{ 483{
822 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
823} 485}
824 486
825unsigned int 487unsigned int
826eio_nready (void) 488eio_nready (void)
827{ 489{
828 return etp_nready (); 490 return etp_nready (EIO_POOL);
829} 491}
830 492
831unsigned int 493unsigned int
832eio_npending (void) 494eio_npending (void)
833{ 495{
834 return etp_npending (); 496 return etp_npending (EIO_POOL);
835} 497}
836 498
837unsigned int ecb_cold 499unsigned int ecb_cold
838eio_nthreads (void) 500eio_nthreads (void)
839{ 501{
840 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
841} 503}
842 504
843void ecb_cold 505void ecb_cold
844eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
845{ 507{
846 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
847} 509}
848 510
849void ecb_cold 511void ecb_cold
850eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
851{ 513{
852 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
853} 515}
854 516
855void ecb_cold 517void ecb_cold
856eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
857{ 519{
858 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
859} 521}
860 522
861void ecb_cold 523void ecb_cold
862eio_set_idle_timeout (unsigned int seconds) 524eio_set_idle_timeout (unsigned int seconds)
863{ 525{
864 etp_set_idle_timeout (seconds); 526 etp_set_idle_timeout (EIO_POOL, seconds);
865} 527}
866 528
867void ecb_cold 529void ecb_cold
868eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
869{ 531{
870 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
871} 533}
872 534
873void ecb_cold 535void ecb_cold
874eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
875{ 537{
876 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
877} 539}
878 540
879int eio_poll (void) 541int eio_poll (void)
880{ 542{
881 return etp_poll (); 543 return etp_poll (EIO_POOL);
882} 544}
883 545
884/*****************************************************************************/ 546/*****************************************************************************/
885/* work around various missing functions */ 547/* work around various missing functions */
886
887#if !HAVE_PREADWRITE
888# undef pread
889# undef pwrite
890# define pread eio__pread
891# define pwrite eio__pwrite
892
893static eio_ssize_t
894eio__pread (int fd, void *buf, size_t count, off_t offset)
895{
896 eio_ssize_t res;
897 off_t ooffset;
898
899 X_LOCK (preadwritelock);
900 ooffset = lseek (fd, 0, SEEK_CUR);
901 lseek (fd, offset, SEEK_SET);
902 res = read (fd, buf, count);
903 lseek (fd, ooffset, SEEK_SET);
904 X_UNLOCK (preadwritelock);
905
906 return res;
907}
908
909static eio_ssize_t
910eio__pwrite (int fd, void *buf, size_t count, off_t offset)
911{
912 eio_ssize_t res;
913 off_t ooffset;
914
915 X_LOCK (preadwritelock);
916 ooffset = lseek (fd, 0, SEEK_CUR);
917 lseek (fd, offset, SEEK_SET);
918 res = write (fd, buf, count);
919 lseek (fd, ooffset, SEEK_SET);
920 X_UNLOCK (preadwritelock);
921
922 return res;
923}
924#endif
925 548
926#ifndef HAVE_UTIMES 549#ifndef HAVE_UTIMES
927 550
928# undef utimes 551# undef utimes
929# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
962 585
963#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
964# undef fdatasync 587# undef fdatasync
965# define fdatasync(fd) fsync (fd) 588# define fdatasync(fd) fsync (fd)
966#endif 589#endif
590
591static int
592eio__syncfs (int fd)
593{
594 int res;
595
596#if HAVE_SYS_SYNCFS
597 res = (int)syscall (__NR_syncfs, (int)(fd));
598#else
599 res = EIO_ENOSYS ();
600#endif
601
602 if (res < 0 && errno == ENOSYS && fd >= 0)
603 sync ();
604
605 return res;
606}
967 607
968/* sync_file_range always needs emulation */ 608/* sync_file_range always needs emulation */
969static int 609static int
970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 610eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
971{ 611{
994} 634}
995 635
996static int 636static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len) 637eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{ 638{
999#if HAVE_FALLOCATE 639#if HAVE_LINUX_FALLOCATE
1000 return fallocate (fd, mode, offset, len); 640 return fallocate (fd, mode, offset, len);
1001#else 641#else
1002 errno = ENOSYS; 642 return EIO_ENOSYS ();
1003 return -1;
1004#endif 643#endif
1005} 644}
1006 645
1007#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
1008# undef readahead 647# undef readahead
1021 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
1022 offset += len; 661 offset += len;
1023 todo -= len; 662 todo -= len;
1024 } 663 }
1025 664
1026 errno = 0; 665 FUBd;
666
667 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
668 /* but not for e.g. EIO or eof, so we also never fail */
1027 return count; 669 return 0;
1028} 670}
1029 671
1030#endif 672#endif
1031 673
1032/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
1033static eio_ssize_t 675static eio_ssize_t
1034eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 676eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1035{ 677{
1036 eio_ssize_t written = 0; 678 eio_ssize_t written = 0;
1037 eio_ssize_t res; 679 eio_ssize_t res;
1038 680
1039 if (!count) 681 if (!count)
1067 709
1068 /* according to source inspection, this is correct, and useful behaviour */ 710 /* according to source inspection, this is correct, and useful behaviour */
1069 if (sbytes) 711 if (sbytes)
1070 res = sbytes; 712 res = sbytes;
1071 713
1072# elif defined (__APPLE__) 714# elif defined __APPLE__
1073 off_t sbytes = count; 715 off_t sbytes = count;
1074 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1075 717
1076 /* according to the manpage, sbytes is always valid */ 718 /* according to the manpage, sbytes is always valid */
1077 if (sbytes) 719 if (sbytes)
1104 HANDLE h = TO_SOCKET (ifd); 746 HANDLE h = TO_SOCKET (ifd);
1105 SetFilePointer (h, offset, 0, FILE_BEGIN); 747 SetFilePointer (h, offset, 0, FILE_BEGIN);
1106 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1107 749
1108#else 750#else
1109 res = -1; 751 res = EIO_ENOSYS ();
1110 errno = ENOSYS;
1111#endif 752#endif
1112 753
1113 /* we assume sendfile can copy at least 128mb in one go */ 754 /* we assume sendfile can copy at least 128mb in one go */
1114 if (res <= 128 * 1024 * 1024) 755 if (res <= 128 * 1024 * 1024)
1115 { 756 {
1175 816
1176 offset += cnt; 817 offset += cnt;
1177 res += cnt; 818 res += cnt;
1178 count -= cnt; 819 count -= cnt;
1179 } 820 }
821
822 FUBd;
1180 } 823 }
1181 824
1182 return res; 825 return res;
1183} 826}
1184 827
1300 943
1301/*****************************************************************************/ 944/*****************************************************************************/
1302/* requests implemented outside eio_execute, because they are so large */ 945/* requests implemented outside eio_execute, because they are so large */
1303 946
1304static void 947static void
1305eio__realpath (eio_req *req, etp_worker *self) 948eio__lseek (eio_req *req)
1306{ 949{
1307 char *rel = req->ptr1; 950 /* this usually gets optimised away completely, or your compiler sucks, */
951 /* or the whence constants really are not 0, 1, 2 */
952 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
953 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
954 : req->int2 == EIO_SEEK_END ? SEEK_END
955 : req->int2;
956
957 req->offs = lseek (req->int1, req->offs, whence);
958 req->result = req->offs == (off_t)-1 ? -1 : 0;
959}
960
961/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
962static int
963eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
964{
1308 char *res; 965 char *res;
966 const char *rel = path;
1309 char *tmp1, *tmp2; 967 char *tmp1, *tmp2;
1310#if SYMLOOP_MAX > 32 968#if SYMLOOP_MAX > 32
1311 int symlinks = SYMLOOP_MAX; 969 int symlinks = SYMLOOP_MAX;
1312#else 970#else
1313 int symlinks = 32; 971 int symlinks = 32;
1314#endif 972#endif
1315 973
1316 req->result = -1;
1317
1318 errno = EINVAL; 974 errno = EINVAL;
1319 if (!rel) 975 if (!rel)
1320 return; 976 return -1;
1321 977
1322 errno = ENOENT; 978 errno = ENOENT;
1323 if (!*rel) 979 if (!*rel)
1324 return; 980 return -1;
1325 981
1326 if (!req->ptr2) 982 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
1327 { 983#ifdef _WIN32
1328 X_LOCK (wrklock); 984 if (_access (rel, 4) != 0)
1329 req->flags |= EIO_FLAG_PTR2_FREE;
1330 X_UNLOCK (wrklock);
1331 req->ptr2 = malloc (PATH_MAX * 3);
1332
1333 errno = ENOMEM;
1334 if (!req->ptr2)
1335 return; 985 return -1;
1336 }
1337 986
1338 res = req->ptr2; 987 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
988
989 errno = ENAMETOOLONG;
990 if (symlinks >= PATH_MAX * 3)
991 return -1;
992
993 errno = EIO;
994 if (symlinks <= 0)
995 return -1;
996
997 return symlinks;
998
999#else
1339 tmp1 = res + PATH_MAX; 1000 tmp1 = res + PATH_MAX;
1340 tmp2 = tmp1 + PATH_MAX; 1001 tmp2 = tmp1 + PATH_MAX;
1341 1002
1342#if 0 /* disabled, the musl way to do things is just too racy */ 1003#if 0 /* disabled, the musl way to do things is just too racy */
1343#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1004#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1347 1008
1348 if (fd >= 0) 1009 if (fd >= 0)
1349 { 1010 {
1350 sprintf (tmp1, "/proc/self/fd/%d", fd); 1011 sprintf (tmp1, "/proc/self/fd/%d", fd);
1351 req->result = readlink (tmp1, res, PATH_MAX); 1012 req->result = readlink (tmp1, res, PATH_MAX);
1013 /* here we should probably stat the open file and the disk file, to make sure they still match */
1352 close (fd); 1014 close (fd);
1353
1354 /* here we should probably stat the open file and the disk file, to make sure they still match */
1355 1015
1356 if (req->result > 0) 1016 if (req->result > 0)
1357 goto done; 1017 goto done;
1358 } 1018 }
1359 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1019 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1360 return; 1020 return -1;
1361 } 1021 }
1362#endif 1022#endif
1363#endif 1023#endif
1364 1024
1365 if (*rel != '/') 1025 if (*rel != '/')
1366 { 1026 {
1027 int len;
1028
1029 errno = ENOENT;
1030 if (wd == EIO_INVALID_WD)
1031 return -1;
1032
1033 if (wd == EIO_CWD)
1034 {
1367 if (!getcwd (res, PATH_MAX)) 1035 if (!getcwd (res, PATH_MAX))
1368 return; 1036 return -1;
1037
1038 len = strlen (res);
1039 }
1040 else
1041 memcpy (res, wd->str, len = wd->len);
1369 1042
1370 if (res [1]) /* only use if not / */ 1043 if (res [1]) /* only use if not / */
1371 res += strlen (res); 1044 res += len;
1372 } 1045 }
1373 1046
1374 while (*rel) 1047 while (*rel)
1375 { 1048 {
1376 eio_ssize_t len, linklen; 1049 eio_ssize_t len, linklen;
1377 char *beg = rel; 1050 const char *beg = rel;
1378 1051
1379 while (*rel && *rel != '/') 1052 while (*rel && *rel != '/')
1380 ++rel; 1053 ++rel;
1381 1054
1382 len = rel - beg; 1055 len = rel - beg;
1394 1067
1395 if (beg [1] == '.' && len == 2) 1068 if (beg [1] == '.' && len == 2)
1396 { 1069 {
1397 /* .. - back up one component, if possible */ 1070 /* .. - back up one component, if possible */
1398 1071
1399 while (res != req->ptr2) 1072 while (res != tmpbuf->ptr)
1400 if (*--res == '/') 1073 if (*--res == '/')
1401 break; 1074 break;
1402 1075
1403 continue; 1076 continue;
1404 } 1077 }
1405 } 1078 }
1406 1079
1407 errno = ENAMETOOLONG; 1080 errno = ENAMETOOLONG;
1408 if (res + 1 + len + 1 >= tmp1) 1081 if (res + 1 + len + 1 >= tmp1)
1409 return; 1082 return -1;
1410 1083
1411 /* copy one component */ 1084 /* copy one component */
1412 *res = '/'; 1085 *res = '/';
1413 memcpy (res + 1, beg, len); 1086 memcpy (res + 1, beg, len);
1414 1087
1415 /* zero-terminate, for readlink */ 1088 /* zero-terminate, for readlink */
1416 res [len + 1] = 0; 1089 res [len + 1] = 0;
1417 1090
1418 /* now check if it's a symlink */ 1091 /* now check if it's a symlink */
1419 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1092 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1420 1093
1421 if (linklen < 0) 1094 if (linklen < 0)
1422 { 1095 {
1423 if (errno != EINVAL) 1096 if (errno != EINVAL)
1424 return; 1097 return -1;
1425 1098
1426 /* it's a normal directory. hopefully */ 1099 /* it's a normal directory. hopefully */
1427 res += len + 1; 1100 res += len + 1;
1428 } 1101 }
1429 else 1102 else
1431 /* yay, it was a symlink - build new path in tmp2 */ 1104 /* yay, it was a symlink - build new path in tmp2 */
1432 int rellen = strlen (rel); 1105 int rellen = strlen (rel);
1433 1106
1434 errno = ENAMETOOLONG; 1107 errno = ENAMETOOLONG;
1435 if (linklen + 1 + rellen >= PATH_MAX) 1108 if (linklen + 1 + rellen >= PATH_MAX)
1436 return; 1109 return -1;
1437 1110
1438 errno = ELOOP; 1111 errno = ELOOP;
1439 if (!--symlinks) 1112 if (!--symlinks)
1440 return; 1113 return -1;
1441 1114
1442 if (*tmp1 == '/') 1115 if (*tmp1 == '/')
1443 res = req->ptr2; /* symlink resolves to an absolute path */ 1116 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1444 1117
1445 /* we need to be careful, as rel might point into tmp2 already */ 1118 /* we need to be careful, as rel might point into tmp2 already */
1446 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1119 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1447 tmp2 [linklen] = '/'; 1120 tmp2 [linklen] = '/';
1448 memcpy (tmp2, tmp1, linklen); 1121 memcpy (tmp2, tmp1, linklen);
1450 rel = tmp2; 1123 rel = tmp2;
1451 } 1124 }
1452 } 1125 }
1453 1126
1454 /* special case for the lone root path */ 1127 /* special case for the lone root path */
1455 if (res == req->ptr2) 1128 if (res == tmpbuf->ptr)
1456 *res++ = '/'; 1129 *res++ = '/';
1457 1130
1458 req->result = res - (char *)req->ptr2; 1131 return res - (char *)tmpbuf->ptr;
1459 1132#endif
1460done:
1461 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1462} 1133}
1463 1134
1464static signed char 1135static signed char
1465eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1466{ 1137{
1651 if (!(flags & EIO_READDIR_DENTS)) 1322 if (!(flags & EIO_READDIR_DENTS))
1652 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1653 1324
1654#ifdef _WIN32 1325#ifdef _WIN32
1655 { 1326 {
1327 int len = strlen ((const char *)req->ptr1);
1656 char *path = malloc (MAX_PATH); 1328 char *path = malloc (MAX_PATH);
1657 _snprintf (path, MAX_PATH, "%s/*", (const char *)req->ptr1); 1329 const char *fmt;
1330 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1331
1332 if (!len)
1333 fmt = "./*";
1334 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1335 fmt = "%s*";
1336 else
1337 fmt = "%s/*";
1338
1339 _snprintf (path, MAX_PATH, fmt, reqpath);
1658 dirp = FindFirstFile (path, &entp); 1340 dirp = FindFirstFile (path, &entp);
1659 free (path); 1341 free (path);
1660 1342
1661 if (!dirp) 1343 if (dirp == INVALID_HANDLE_VALUE)
1662 { 1344 {
1345 /* should steal _dosmaperr */
1663 switch (GetLastError ()) 1346 switch (GetLastError ())
1664 { 1347 {
1665 case ERROR_FILE_NOT_FOUND: 1348 case ERROR_FILE_NOT_FOUND:
1666 req->result = 0; 1349 req->result = 0;
1667 break; 1350 break;
1668 1351
1352 case ERROR_INVALID_NAME:
1669 case ERROR_PATH_NOT_FOUND: 1353 case ERROR_PATH_NOT_FOUND:
1670 case ERROR_NO_MORE_FILES: 1354 case ERROR_NO_MORE_FILES:
1671 errno = ENOENT; 1355 errno = ENOENT;
1672 break; 1356 break;
1673 1357
1674 case ERROR_NOT_ENOUGH_MEMORY: 1358 case ERROR_NOT_ENOUGH_MEMORY:
1675 errno = ENOMEM; 1359 errno = ENOMEM;
1676 break; 1360 break;
1677 1361
1678 default: 1362 default:
1679 errno = EINVAL; 1363 errno = EINVAL;
1680 break; 1364 break;
1681 } 1365 }
1682 1366
1367 return;
1683 } 1368 }
1684 } 1369 }
1685#else 1370#else
1371 #if HAVE_AT
1372 if (req->wd)
1373 {
1374 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1375
1376 if (fd < 0)
1377 return;
1378
1379 dirp = fdopendir (fd);
1380
1381 if (!dirp)
1382 close (fd);
1383 }
1384 else
1686 dirp = opendir (req->ptr1); 1385 dirp = opendir (req->ptr1);
1386 #else
1387 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1388 #endif
1389
1390 if (!dirp)
1391 return;
1687#endif 1392#endif
1688 1393
1689 if (req->flags & EIO_FLAG_PTR1_FREE) 1394 if (req->flags & EIO_FLAG_PTR1_FREE)
1690 free (req->ptr1); 1395 free (req->ptr1);
1691 1396
1692 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1693 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1694 req->ptr2 = names = malloc (namesalloc); 1399 req->ptr2 = names = malloc (namesalloc);
1695 1400
1696 if (dirp && names && (!flags || dents)) 1401 if (!names || (flags && !dents))
1402 return;
1403
1697 for (;;) 1404 for (;;)
1698 { 1405 {
1699 int more; 1406 int done;
1700 1407
1701#ifdef _WIN32 1408#ifdef _WIN32
1702 more = dirp; 1409 done = !dirp;
1703#else 1410#else
1704 errno = 0; 1411 errno = 0;
1705 entp = readdir (dirp); 1412 entp = readdir (dirp);
1706 more = entp; 1413 done = !entp;
1707#endif 1414#endif
1708 1415
1709 if (!more) 1416 if (done)
1710 { 1417 {
1711#ifndef _WIN32 1418#ifndef _WIN32
1712 int old_errno = errno; 1419 int old_errno = errno;
1713 closedir (dirp); 1420 closedir (dirp);
1714 errno = old_errno; 1421 errno = old_errno;
1715 1422
1716 if (errno) 1423 if (errno)
1717 break; 1424 break;
1718#endif 1425#endif
1719 1426
1720 /* sort etc. */ 1427 /* sort etc. */
1721 req->int1 = flags; 1428 req->int1 = flags;
1722 req->result = dentoffs; 1429 req->result = dentoffs;
1723 1430
1724 if (flags & EIO_READDIR_STAT_ORDER) 1431 if (flags & EIO_READDIR_STAT_ORDER)
1725 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1432 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1726 else if (flags & EIO_READDIR_DIRS_FIRST) 1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1727 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1728 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1435 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1729 else 1436 else
1437 {
1438 /* in this case, all is known, and we just put dirs first and sort them */
1439 eio_dirent *oth = dents + dentoffs;
1440 eio_dirent *dir = dents;
1441
1442 /* now partition dirs to the front, and non-dirs to the back */
1443 /* by walking from both sides and swapping if necessary */
1444 while (oth > dir)
1445 {
1446 if (dir->type == EIO_DT_DIR)
1447 ++dir;
1448 else if ((--oth)->type == EIO_DT_DIR)
1449 {
1450 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1451
1452 ++dir;
1453 }
1454 }
1455
1456 /* now sort the dirs only (dirs all have the same score) */
1457 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1458 }
1459
1460 break;
1461 }
1462
1463 /* now add the entry to our list(s) */
1464 name = D_NAME (entp);
1465
1466 /* skip . and .. entries */
1467 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1468 {
1469 int len = D_NAMLEN (entp) + 1;
1470
1471 while (ecb_expect_false (namesoffs + len > namesalloc))
1472 {
1473 namesalloc *= 2;
1474 req->ptr2 = names = realloc (names, namesalloc);
1475
1476 if (!names)
1477 break;
1478 }
1479
1480 memcpy (names + namesoffs, name, len);
1481
1482 if (dents)
1483 {
1484 struct eio_dirent *ent;
1485
1486 if (ecb_expect_false (dentoffs == dentalloc))
1730 { 1487 {
1731 /* in this case, all is known, and we just put dirs first and sort them */ 1488 dentalloc *= 2;
1489 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1490
1491 if (!dents)
1492 break;
1493 }
1494
1732 eio_dirent *oth = dents + dentoffs; 1495 ent = dents + dentoffs;
1733 eio_dirent *dir = dents;
1734 1496
1735 /* now partition dirs to the front, and non-dirs to the back */ 1497 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1736 /* by walking from both sides and swapping if necessary */ 1498 ent->namelen = len - 1;
1737 while (oth > dir) 1499 ent->inode = D_INO (entp);
1500
1501 inode_bits |= ent->inode;
1502
1503 switch (D_TYPE (entp))
1504 {
1505 default:
1506 ent->type = EIO_DT_UNKNOWN;
1507 flags |= EIO_READDIR_FOUND_UNKNOWN;
1508 break;
1509
1510 #ifdef DT_FIFO
1511 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1512 #endif
1513 #ifdef DT_CHR
1514 case DT_CHR: ent->type = EIO_DT_CHR; break;
1515 #endif
1516 #ifdef DT_MPC
1517 case DT_MPC: ent->type = EIO_DT_MPC; break;
1518 #endif
1519 #ifdef DT_DIR
1520 case DT_DIR: ent->type = EIO_DT_DIR; break;
1521 #endif
1522 #ifdef DT_NAM
1523 case DT_NAM: ent->type = EIO_DT_NAM; break;
1524 #endif
1525 #ifdef DT_BLK
1526 case DT_BLK: ent->type = EIO_DT_BLK; break;
1527 #endif
1528 #ifdef DT_MPB
1529 case DT_MPB: ent->type = EIO_DT_MPB; break;
1530 #endif
1531 #ifdef DT_REG
1532 case DT_REG: ent->type = EIO_DT_REG; break;
1533 #endif
1534 #ifdef DT_NWK
1535 case DT_NWK: ent->type = EIO_DT_NWK; break;
1536 #endif
1537 #ifdef DT_CMP
1538 case DT_CMP: ent->type = EIO_DT_CMP; break;
1539 #endif
1540 #ifdef DT_LNK
1541 case DT_LNK: ent->type = EIO_DT_LNK; break;
1542 #endif
1543 #ifdef DT_SOCK
1544 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1545 #endif
1546 #ifdef DT_DOOR
1547 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1548 #endif
1549 #ifdef DT_WHT
1550 case DT_WHT: ent->type = EIO_DT_WHT; break;
1551 #endif
1552 }
1553
1554 ent->score = 7;
1555
1556 if (flags & EIO_READDIR_DIRS_FIRST)
1557 {
1558 if (ent->type == EIO_DT_UNKNOWN)
1738 { 1559 {
1739 if (dir->type == EIO_DT_DIR) 1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1740 ++dir; 1561 ent->score = 1;
1741 else if ((--oth)->type == EIO_DT_DIR) 1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1742 { 1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1743 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1744
1745 ++dir;
1746 }
1747 } 1564 }
1748 1565 else if (ent->type == EIO_DT_DIR)
1749 /* now sort the dirs only (dirs all have the same score) */ 1566 ent->score = 0;
1750 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1751 } 1567 }
1752
1753 break;
1754 }
1755
1756 /* now add the entry to our list(s) */
1757 name = D_NAME (entp);
1758
1759 /* skip . and .. entries */
1760 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1761 {
1762 int len = D_NAMLEN (entp) + 1;
1763
1764 while (ecb_expect_false (namesoffs + len > namesalloc))
1765 {
1766 namesalloc *= 2;
1767 req->ptr2 = names = realloc (names, namesalloc);
1768
1769 if (!names)
1770 break;
1771 } 1568 }
1772 1569
1773 memcpy (names + namesoffs, name, len);
1774
1775 if (dents)
1776 {
1777 struct eio_dirent *ent;
1778
1779 if (ecb_expect_false (dentoffs == dentalloc))
1780 {
1781 dentalloc *= 2;
1782 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1783
1784 if (!dents)
1785 break;
1786 }
1787
1788 ent = dents + dentoffs;
1789
1790 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1791 ent->namelen = len - 1;
1792 ent->inode = D_INO (entp);
1793
1794 inode_bits |= ent->inode;
1795
1796 switch (D_TYPE (entp))
1797 {
1798 default:
1799 ent->type = EIO_DT_UNKNOWN;
1800 flags |= EIO_READDIR_FOUND_UNKNOWN;
1801 break;
1802
1803 #ifdef DT_FIFO
1804 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1805 #endif
1806 #ifdef DT_CHR
1807 case DT_CHR: ent->type = EIO_DT_CHR; break;
1808 #endif
1809 #ifdef DT_MPC
1810 case DT_MPC: ent->type = EIO_DT_MPC; break;
1811 #endif
1812 #ifdef DT_DIR
1813 case DT_DIR: ent->type = EIO_DT_DIR; break;
1814 #endif
1815 #ifdef DT_NAM
1816 case DT_NAM: ent->type = EIO_DT_NAM; break;
1817 #endif
1818 #ifdef DT_BLK
1819 case DT_BLK: ent->type = EIO_DT_BLK; break;
1820 #endif
1821 #ifdef DT_MPB
1822 case DT_MPB: ent->type = EIO_DT_MPB; break;
1823 #endif
1824 #ifdef DT_REG
1825 case DT_REG: ent->type = EIO_DT_REG; break;
1826 #endif
1827 #ifdef DT_NWK
1828 case DT_NWK: ent->type = EIO_DT_NWK; break;
1829 #endif
1830 #ifdef DT_CMP
1831 case DT_CMP: ent->type = EIO_DT_CMP; break;
1832 #endif
1833 #ifdef DT_LNK
1834 case DT_LNK: ent->type = EIO_DT_LNK; break;
1835 #endif
1836 #ifdef DT_SOCK
1837 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1838 #endif
1839 #ifdef DT_DOOR
1840 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1841 #endif
1842 #ifdef DT_WHT
1843 case DT_WHT: ent->type = EIO_DT_WHT; break;
1844 #endif
1845 }
1846
1847 ent->score = 7;
1848
1849 if (flags & EIO_READDIR_DIRS_FIRST)
1850 {
1851 if (ent->type == EIO_DT_UNKNOWN)
1852 {
1853 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1854 ent->score = 1;
1855 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1856 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1857 }
1858 else if (ent->type == EIO_DT_DIR)
1859 ent->score = 0;
1860 }
1861 }
1862
1863 namesoffs += len; 1570 namesoffs += len;
1864 ++dentoffs; 1571 ++dentoffs;
1865 } 1572 }
1866 1573
1867 if (EIO_CANCELLED (req)) 1574 if (EIO_CANCELLED (req))
1868 { 1575 {
1869 errno = ECANCELED; 1576 errno = ECANCELED;
1870 break; 1577 break;
1871 } 1578 }
1872 1579
1873#ifdef _WIN32 1580#ifdef _WIN32
1874 if (!FindNextFile (dirp, &entp)) 1581 if (!FindNextFile (dirp, &entp))
1875 { 1582 {
1876 FindClose (dirp); 1583 FindClose (dirp);
1877 dirp = 0; 1584 dirp = 0;
1878 } 1585 }
1879#endif 1586#endif
1880 } 1587 }
1881} 1588}
1589
1590/*****************************************************************************/
1591/* working directory stuff */
1592/* various deficiencies in the posix 2008 api force us to */
1593/* keep the absolute path in string form at all times */
1594/* fuck yeah. */
1595
1596#if !HAVE_AT
1597
1598/* a bit like realpath, but usually faster because it doesn'T have to return */
1599/* an absolute or canonical path */
1600static const char *
1601wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1602{
1603 if (!wd || *path == '/')
1604 return path;
1605
1606 if (path [0] == '.' && !path [1])
1607 return wd->str;
1608
1609 {
1610 int l1 = wd->len;
1611 int l2 = strlen (path);
1612
1613 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1614
1615 memcpy (res, wd->str, l1);
1616 res [l1] = '/';
1617 memcpy (res + l1 + 1, path, l2 + 1);
1618
1619 return res;
1620 }
1621}
1622
1623#endif
1624
1625static eio_wd
1626eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1627{
1628 int fd;
1629 eio_wd res;
1630 int len = eio__realpath (tmpbuf, wd, path);
1631
1632 if (len < 0)
1633 return EIO_INVALID_WD;
1634
1635#if HAVE_AT
1636 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1637
1638 if (fd < 0)
1639 return EIO_INVALID_WD;
1640#endif
1641
1642 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1643
1644#if HAVE_AT
1645 res->fd = fd;
1646#endif
1647
1648 res->len = len;
1649 memcpy (res->str, tmpbuf->ptr, len);
1650 res->str [len] = 0;
1651
1652 return res;
1653}
1654
1655eio_wd
1656eio_wd_open_sync (eio_wd wd, const char *path)
1657{
1658 struct etp_tmpbuf tmpbuf = { };
1659 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1660 free (tmpbuf.ptr);
1661
1662 return wd;
1663}
1664
1665void
1666eio_wd_close_sync (eio_wd wd)
1667{
1668 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1669 {
1670 #if HAVE_AT
1671 close (wd->fd);
1672 #endif
1673 free (wd);
1674 }
1675}
1676
1677#if HAVE_AT
1678
1679/* they forgot these */
1680
1681static int
1682eio__truncateat (int dirfd, const char *path, off_t length)
1683{
1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1685 int res;
1686
1687 if (fd < 0)
1688 return fd;
1689
1690 res = ftruncate (fd, length);
1691 close (fd);
1692 return res;
1693}
1694
1695static int
1696eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1697{
1698 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1699 int res;
1700
1701 if (fd < 0)
1702 return fd;
1703
1704 res = fstatvfs (fd, buf);
1705 close (fd);
1706 return res;
1707
1708}
1709
1710#endif
1882 1711
1883/*****************************************************************************/ 1712/*****************************************************************************/
1884 1713
1885#define ALLOC(len) \ 1714#define ALLOC(len) \
1886 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1887 { \ 1716 { \
1888 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1889 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1890 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1891 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1892 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1893 { \ 1722 { \
1894 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1895 req->result = -1; \ 1724 req->result = -1; \
1896 break; \ 1725 break; \
1897 } \ 1726 } \
1898 } 1727 }
1899 1728
1900X_THREAD_PROC (etp_proc)
1901{
1902 ETP_REQ *req;
1903 struct timespec ts;
1904 etp_worker *self = (etp_worker *)thr_arg;
1905
1906 /* try to distribute timeouts somewhat evenly */
1907 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1908
1909 for (;;)
1910 {
1911 ts.tv_sec = 0;
1912
1913 X_LOCK (reqlock);
1914
1915 for (;;)
1916 {
1917 self->req = req = reqq_shift (&req_queue);
1918
1919 if (req)
1920 break;
1921
1922 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1923 {
1924 X_UNLOCK (reqlock);
1925 X_LOCK (wrklock);
1926 --started;
1927 X_UNLOCK (wrklock);
1928 goto quit;
1929 }
1930
1931 ++idle;
1932
1933 if (idle <= max_idle)
1934 /* we are allowed to idle, so do so without any timeout */
1935 X_COND_WAIT (reqwait, reqlock);
1936 else
1937 {
1938 /* initialise timeout once */
1939 if (!ts.tv_sec)
1940 ts.tv_sec = time (0) + idle_timeout;
1941
1942 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1943 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1944 }
1945
1946 --idle;
1947 }
1948
1949 --nready;
1950
1951 X_UNLOCK (reqlock);
1952
1953 if (req->type < 0)
1954 goto quit;
1955
1956 ETP_EXECUTE (self, req);
1957
1958 X_LOCK (reslock);
1959
1960 ++npending;
1961
1962 if (!reqq_push (&res_queue, req) && want_poll_cb)
1963 want_poll_cb ();
1964
1965 self->req = 0;
1966 etp_worker_clear (self);
1967
1968 X_UNLOCK (reslock);
1969 }
1970
1971quit:
1972 X_LOCK (wrklock);
1973 etp_worker_free (self);
1974 X_UNLOCK (wrklock);
1975
1976 return 0;
1977}
1978
1979/*****************************************************************************/ 1729/*****************************************************************************/
1980 1730
1981int ecb_cold 1731int ecb_cold
1982eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1983{ 1733{
1984 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1735 eio_done_poll_cb = done_poll;
1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1985} 1738}
1986 1739
1987ecb_inline void 1740ecb_inline void
1988eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1989{ 1742{
1990 free (req); 1743 free (req);
1991} 1744}
1992 1745
1993#define REQ(rtype) \ 1746#define REQ(rtype) \
1994 eio_req *req; \ 1747 eio_req *req; \
1995 \ 1748 \
1996 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1997 if (!req) \ 1750 if (!req) \
1998 return 0; \ 1751 return 0; \
2012 { \ 1765 { \
2013 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
2014 return 0; \ 1767 return 0; \
2015 } 1768 }
2016 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
2017static void 1772static void
2018eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
2019{ 1774{
1775#if HAVE_AT
1776 int dirfd;
1777#else
1778 const char *path;
1779#endif
1780
2020 if (ecb_expect_false (EIO_CANCELLED (req))) 1781 if (ecb_expect_false (EIO_CANCELLED (req)))
2021 { 1782 {
2022 req->result = -1; 1783 req->result = -1;
2023 req->errorno = ECANCELED; 1784 req->errorno = ECANCELED;
2024 return; 1785 return;
2025 } 1786 }
2026 1787
1788 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1789 {
1790 req->result = -1;
1791 req->errorno = ENOENT;
1792 return;
1793 }
1794
1795 if (req->type >= EIO_OPEN)
1796 {
1797 #if HAVE_AT
1798 dirfd = WD2FD (req->wd);
1799 #else
1800 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1801 #endif
1802 }
1803
2027 switch (req->type) 1804 switch (req->type)
2028 { 1805 {
1806 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1807 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1808 break;
1809 case EIO_WD_CLOSE: req->result = 0;
1810 eio_wd_close_sync (req->wd); break;
1811
1812 case EIO_SEEK: eio__lseek (req); break;
2029 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
2030 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
2031 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
2032 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
2033 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
2034 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2035 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
2036 1820
2037 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1821 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2038 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1822 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1823
1824#if HAVE_AT
2039 1825
2040 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2041 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1827 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2042 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1834
1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1848 case EIO_READLINK: ALLOC (PATH_MAX);
1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1852 case EIO_UTIME:
1853 case EIO_FUTIME:
1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1876#else
1877
1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2043 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1886
1887 case EIO_UNLINK: req->result = unlink (path ); break;
1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1894 case EIO_READLINK: ALLOC (PATH_MAX);
1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1898
1899 case EIO_UTIME:
1900 case EIO_FUTIME:
1901 {
1902 struct timeval tv[2];
1903 struct timeval *times;
1904
1905 if (req->nv1 != -1. || req->nv2 != -1.)
1906 {
1907 tv[0].tv_sec = req->nv1;
1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1909 tv[1].tv_sec = req->nv2;
1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1911
1912 times = tv;
1913 }
1914 else
1915 times = 0;
1916
1917 req->result = req->type == EIO_FUTIME
1918 ? futimes (req->int1, times)
1919 : utimes (req->ptr1, times);
1920 }
1921 break;
1922
1923#endif
1924
1925 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1926 {
1927 ALLOC (req->result);
1928 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1929 }
1930 break;
1931
2044 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2045 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2046 1934
2047 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2048 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2049 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2050 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2051 1937
2052 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2053 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2054 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2055 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2056 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2057 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2058 1941
2059 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2060 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
2061 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2062 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2063 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2064 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2065 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2066 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2067 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2068 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2069
2070 case EIO_REALPATH: eio__realpath (req, self); break;
2071
2072 case EIO_READLINK: ALLOC (PATH_MAX);
2073 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2074
2075 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
2076 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
2077 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1946 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1947 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1948 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2078 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2079 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2080 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2081 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2082 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2083 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2084 1954
2085 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
2086 1956
2087 case EIO_BUSY: 1957 case EIO_BUSY:
2097 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
2098 } 1968 }
2099#endif 1969#endif
2100 break; 1970 break;
2101 1971
2102 case EIO_UTIME: 1972#if 0
2103 case EIO_FUTIME:
2104 {
2105 struct timeval tv[2];
2106 struct timeval *times;
2107
2108 if (req->nv1 != -1. || req->nv2 != -1.)
2109 {
2110 tv[0].tv_sec = req->nv1;
2111 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2112 tv[1].tv_sec = req->nv2;
2113 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2114
2115 times = tv;
2116 }
2117 else
2118 times = 0;
2119
2120 req->result = req->type == EIO_FUTIME
2121 ? futimes (req->int1, times)
2122 : utimes (req->ptr1, times);
2123 }
2124 break;
2125
2126 case EIO_GROUP: 1973 case EIO_GROUP:
2127 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
2128 1976
2129 case EIO_NOP: 1977 case EIO_NOP:
2130 req->result = 0; 1978 req->result = 0;
2131 break; 1979 break;
2132 1980
2133 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
2134 req->feed (req); 1982 req->feed (req);
2135 break; 1983 break;
2136 1984
2137 default: 1985 default:
2138 errno = ENOSYS;
2139 req->result = -1; 1986 req->result = EIO_ENOSYS ();
2140 break; 1987 break;
2141 } 1988 }
2142 1989
2143 req->errorno = errno; 1990 req->errorno = errno;
2144} 1991}
2145 1992
2146#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
2147 1994
1995eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_WD_OPEN); PATH; SEND;
1998}
1999
2000eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2003}
2004
2148eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
2149{ 2006{
2150 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
2151} 2008}
2152 2009
2168eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2025eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2169{ 2026{
2170 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2027 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2171} 2028}
2172 2029
2030eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2031{
2032 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2033}
2034
2035eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2043}
2044
2173eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2045eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2174{ 2046{
2175 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2047 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2176} 2048}
2177 2049
2183eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2184{ 2056{
2185 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2186} 2058}
2187 2059
2188eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2189{
2190 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2191}
2192
2193eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2194{ 2061{
2195 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2196} 2063}
2197 2064
2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2199{
2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2201}
2202
2203eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2065eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2204{ 2066{
2205 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2206} 2068}
2207 2069
2208eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2070eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2209{ 2071{
2210 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2072 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2073}
2074
2075eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2076{
2077 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2211} 2078}
2212 2079
2213eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2080eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2214{ 2081{
2215 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2082 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2413void 2280void
2414eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
2415{ 2282{
2416 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2283 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2417 2284
2418 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
2419 2286
2420 ++grp->size; 2287 ++grp->size;
2421 req->grp = grp; 2288 req->grp = grp;
2422 2289
2423 req->grp_prev = 0; 2290 req->grp_prev = 0;
2433/* misc garbage */ 2300/* misc garbage */
2434 2301
2435eio_ssize_t 2302eio_ssize_t
2436eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2303eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2437{ 2304{
2438 etp_worker wrk;
2439 eio_ssize_t ret;
2440
2441 wrk.dbuf = 0;
2442
2443 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
2444
2445 if (wrk.dbuf)
2446 free (wrk.dbuf);
2447
2448 return ret;
2449} 2306}
2450 2307

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines