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

Comparing libeio/eio.c (file contents):
Revision 1.99 by root, Tue Jul 26 11:07:08 2011 UTC vs.
Revision 1.148 by root, Tue Feb 20 06:05:19 2018 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) 2007,2008,2009,2010,2011,2012,2013,2016,2017 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,
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
50#include "xthread.h"
51
52#include <errno.h> 47#include <errno.h>
53#include <stddef.h> 48#include <stddef.h>
54#include <stdlib.h> 49#include <stdlib.h>
55#include <string.h> 50#include <string.h>
56#include <errno.h> 51#include <errno.h>
105#ifdef _WIN32 100#ifdef _WIN32
106 101
107 #undef PAGESIZE 102 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */ 103 #define PAGESIZE 4096 /* GetSystemInfo? */
109 104
105 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
110 #ifdef EIO_STRUCT_STATI64 109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf) 111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf) 112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif 113 #endif
114 #define lstat(path,buf) stat (path,buf) 114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118 118
119 #define chmod(path,mode) _chmod (path, mode) 119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd) 120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2) 121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
122 123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
123 #define fchmod(fd,mode) EIO_ENOSYS () 126 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS () 127 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS () 128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS () 131 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS () 132 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS () 133 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS () 134 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS () 135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
133 171
134 /* rename() uses MoveFile, which fails to overwrite */ 172 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu) 173 #define rename(old,neu) eio__rename (old, neu)
136 174
137 static int 175 static int
165 /* we could even stat and see if it exists */ 203 /* we could even stat and see if it exists */
166 static int 204 static int
167 symlink (const char *old, const char *neu) 205 symlink (const char *old, const char *neu)
168 { 206 {
169 #if WINVER >= 0x0600 207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
170 if (CreateSymbolicLink (neu, old, 1)) 215 if (CreateSymbolicLink (neu, old, flags))
171 return 0; 216 return 0;
172
173 if (CreateSymbolicLink (neu, old, 0))
174 return 0;
175 #endif 217 #endif
176 218
177 return EIO_ERRNO (ENOENT, -1); 219 return EIO_ERRNO (ENOENT, -1);
178 } 220 }
179 221
180 /* POSIX API only */ 222 /* POSIX API only, causing trouble for win32 apps */
181 #define CreateHardLink(neu,old,flags) 0 223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
182 #define CreateSymbolicLink(neu,old,flags) 0 224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
183 225
184 struct statvfs 226 struct statvfs
185 { 227 {
186 int dummy; 228 int dummy;
187 }; 229 };
191 #define D_NAME(entp) entp.cFileName 233 #define D_NAME(entp) entp.cFileName
192 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG) 234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
193 235
194#else 236#else
195 237
238 #include <sys/ioctl.h>
196 #include <sys/time.h> 239 #include <sys/time.h>
197 #include <sys/select.h> 240 #include <sys/select.h>
198 #include <sys/statvfs.h>
199 #include <unistd.h> 241 #include <unistd.h>
200 #include <signal.h> 242 #include <signal.h>
201 #include <dirent.h> 243 #include <dirent.h>
202 244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
253
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h> 255 #include <sys/mman.h>
205 #endif 256 #endif
206 257
207 #define D_NAME(entp) entp->d_name 258 #define D_NAME(entp) entp->d_name
208 259
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */ 262 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno 263 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen 264 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino 266 #define D_INO(de) (de)->d_ino
234# include <utime.h> 285# include <utime.h>
235#endif 286#endif
236 287
237#if HAVE_SYS_SYSCALL_H 288#if HAVE_SYS_SYSCALL_H
238# include <sys/syscall.h> 289# include <sys/syscall.h>
239#endif
240
241#if HAVE_SYS_PRCTL_H
242# include <sys/prctl.h>
243#endif 290#endif
244 291
245#if HAVE_SENDFILE 292#if HAVE_SENDFILE
246# if __linux 293# if __linux
247# include <sys/sendfile.h> 294# include <sys/sendfile.h>
255# else 302# else
256# error sendfile support requested but not available 303# error sendfile support requested but not available
257# endif 304# endif
258#endif 305#endif
259 306
307#if HAVE_RENAMEAT2
308# include <sys/syscall.h>
309# include <linux/fs.h>
310#endif
311
260#ifndef D_TYPE 312#ifndef D_TYPE
261# define D_TYPE(de) 0 313# define D_TYPE(de) 0
262#endif 314#endif
263#ifndef D_INO 315#ifndef D_INO
264# define D_INO(de) 0 316# define D_INO(de) 0
272# define NAME_MAX 4096 324# define NAME_MAX 4096
273#endif 325#endif
274 326
275/* used for readlink etc. */ 327/* used for readlink etc. */
276#ifndef PATH_MAX 328#ifndef PATH_MAX
277# define PATH_MAX 4096 329# define PATH_MAX 0
278#endif 330#endif
331
332#ifndef EIO_PATH_MIN
333# define EIO_PATH_MIN 8160
334#endif
335
336#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
279 337
280/* buffer size for various temporary buffers */ 338/* buffer size for various temporary buffers */
281#define EIO_BUFSIZE 65536 339#define EIO_BUFSIZE 65536
282 340
283#define dBUF \ 341#define dBUF \
284 char *eio_buf = malloc (EIO_BUFSIZE); \ 342 char *eio_buf = malloc (EIO_BUFSIZE); \
285 errno = ENOMEM; \ 343 errno = ENOMEM; \
286 if (!eio_buf) \ 344 if (!eio_buf) \
287 return -1 345 return -1
288 346
289#define FUBd \ 347#define FUBd \
290 free (eio_buf) 348 free (eio_buf)
291 349
292#define EIO_TICKS ((1000000 + 1023) >> 10) 350/*****************************************************************************/
351
352struct etp_tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
373
374/*****************************************************************************/
293 375
294#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
295#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
296 378
379#define ETP_TYPE_QUIT -1
380#define ETP_TYPE_GROUP EIO_GROUP
381
382static void eio_nop_callback (void) { }
383static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
297struct etp_worker; 389struct etp_worker;
298
299#define ETP_REQ eio_req 390#define ETP_REQ eio_req
300#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
301static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
302#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
303static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
304#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
305 396
306/*****************************************************************************/ 397#include "etp.c"
307 398
308#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 399static struct etp_pool eio_pool;
309 400#define EIO_POOL (&eio_pool)
310/* calculate time difference in ~1/EIO_TICKS of a second */
311ecb_inline int
312tvdiff (struct timeval *tv1, struct timeval *tv2)
313{
314 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
315 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
316}
317
318static unsigned int started, idle, wanted = 4;
319
320static void (*want_poll_cb) (void);
321static void (*done_poll_cb) (void);
322
323static unsigned int max_poll_time; /* reslock */
324static unsigned int max_poll_reqs; /* reslock */
325
326static volatile unsigned int nreqs; /* reqlock */
327static volatile unsigned int nready; /* reqlock */
328static volatile unsigned int npending; /* reqlock */
329static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
330static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
331
332static xmutex_t wrklock;
333static xmutex_t reslock;
334static xmutex_t reqlock;
335static xcond_t reqwait;
336
337#if !HAVE_PREADWRITE
338/*
339 * make our pread/pwrite emulation safe against themselves, but not against
340 * normal read/write by using a mutex. slows down execution a lot,
341 * but that's your problem, not mine.
342 */
343static xmutex_t preadwritelock;
344#endif
345
346typedef struct etp_worker
347{
348 /* locked by wrklock */
349 struct etp_worker *prev, *next;
350
351 xthread_t tid;
352
353 /* locked by reslock, reqlock or wrklock */
354 ETP_REQ *req; /* currently processed request */
355
356#ifdef ETP_WORKER_COMMON
357 ETP_WORKER_COMMON
358#endif
359} etp_worker;
360
361static etp_worker wrk_first; /* NOT etp */
362
363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
365
366/* worker threads management */
367
368static void ecb_cold
369etp_worker_clear (etp_worker *wrk)
370{
371}
372
373static void ecb_cold
374etp_worker_free (etp_worker *wrk)
375{
376 wrk->next->prev = wrk->prev;
377 wrk->prev->next = wrk->next;
378
379 free (wrk);
380}
381
382static unsigned int
383etp_nreqs (void)
384{
385 int retval;
386 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
387 retval = nreqs;
388 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
389 return retval;
390}
391
392static unsigned int
393etp_nready (void)
394{
395 unsigned int retval;
396
397 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
398 retval = nready;
399 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
400
401 return retval;
402}
403
404static unsigned int
405etp_npending (void)
406{
407 unsigned int retval;
408
409 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
410 retval = npending;
411 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
412
413 return retval;
414}
415
416static unsigned int
417etp_nthreads (void)
418{
419 unsigned int retval;
420
421 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
422 retval = started;
423 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
424
425 return retval;
426}
427
428/*
429 * a somewhat faster data structure might be nice, but
430 * with 8 priorities this actually needs <20 insns
431 * per shift, the most expensive operation.
432 */
433typedef struct {
434 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
435 int size;
436} etp_reqq;
437
438static etp_reqq req_queue;
439static etp_reqq res_queue;
440
441static void ecb_noinline ecb_cold
442reqq_init (etp_reqq *q)
443{
444 int pri;
445
446 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
447 q->qs[pri] = q->qe[pri] = 0;
448
449 q->size = 0;
450}
451
452static int ecb_noinline
453reqq_push (etp_reqq *q, ETP_REQ *req)
454{
455 int pri = req->pri;
456 req->next = 0;
457
458 if (q->qe[pri])
459 {
460 q->qe[pri]->next = req;
461 q->qe[pri] = req;
462 }
463 else
464 q->qe[pri] = q->qs[pri] = req;
465
466 return q->size++;
467}
468
469static ETP_REQ * ecb_noinline
470reqq_shift (etp_reqq *q)
471{
472 int pri;
473
474 if (!q->size)
475 return 0;
476
477 --q->size;
478
479 for (pri = ETP_NUM_PRI; pri--; )
480 {
481 eio_req *req = q->qs[pri];
482
483 if (req)
484 {
485 if (!(q->qs[pri] = (eio_req *)req->next))
486 q->qe[pri] = 0;
487
488 return req;
489 }
490 }
491
492 abort ();
493}
494
495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
497{
498 X_MUTEX_CREATE (wrklock);
499 X_MUTEX_CREATE (reslock);
500 X_MUTEX_CREATE (reqlock);
501 X_COND_CREATE (reqwait);
502
503 reqq_init (&req_queue);
504 reqq_init (&res_queue);
505
506 wrk_first.next =
507 wrk_first.prev = &wrk_first;
508
509 started = 0;
510 idle = 0;
511 nreqs = 0;
512 nready = 0;
513 npending = 0;
514
515 want_poll_cb = want_poll;
516 done_poll_cb = done_poll;
517
518 return 0;
519}
520
521X_THREAD_PROC (etp_proc);
522
523static void ecb_cold
524etp_start_thread (void)
525{
526 etp_worker *wrk = calloc (1, sizeof (etp_worker));
527
528 /*TODO*/
529 assert (("unable to allocate worker thread data", wrk));
530
531 X_LOCK (wrklock);
532
533 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
534 {
535 wrk->prev = &wrk_first;
536 wrk->next = wrk_first.next;
537 wrk_first.next->prev = wrk;
538 wrk_first.next = wrk;
539 ++started;
540 }
541 else
542 free (wrk);
543
544 X_UNLOCK (wrklock);
545}
546
547static void
548etp_maybe_start_thread (void)
549{
550 if (ecb_expect_true (etp_nthreads () >= wanted))
551 return;
552
553 /* todo: maybe use idle here, but might be less exact */
554 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
555 return;
556
557 etp_start_thread ();
558}
559
560static void ecb_cold
561etp_end_thread (void)
562{
563 eio_req *req = calloc (1, sizeof (eio_req));
564
565 req->type = -1;
566 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
567
568 X_LOCK (reqlock);
569 reqq_push (&req_queue, req);
570 X_COND_SIGNAL (reqwait);
571 X_UNLOCK (reqlock);
572
573 X_LOCK (wrklock);
574 --started;
575 X_UNLOCK (wrklock);
576}
577
578static int
579etp_poll (void)
580{
581 unsigned int maxreqs;
582 unsigned int maxtime;
583 struct timeval tv_start, tv_now;
584
585 X_LOCK (reslock);
586 maxreqs = max_poll_reqs;
587 maxtime = max_poll_time;
588 X_UNLOCK (reslock);
589
590 if (maxtime)
591 gettimeofday (&tv_start, 0);
592
593 for (;;)
594 {
595 ETP_REQ *req;
596
597 etp_maybe_start_thread ();
598
599 X_LOCK (reslock);
600 req = reqq_shift (&res_queue);
601
602 if (req)
603 {
604 --npending;
605
606 if (!res_queue.size && done_poll_cb)
607 done_poll_cb ();
608 }
609
610 X_UNLOCK (reslock);
611
612 if (!req)
613 return 0;
614
615 X_LOCK (reqlock);
616 --nreqs;
617 X_UNLOCK (reqlock);
618
619 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
620 {
621 req->int1 = 1; /* mark request as delayed */
622 continue;
623 }
624 else
625 {
626 int res = ETP_FINISH (req);
627 if (ecb_expect_false (res))
628 return res;
629 }
630
631 if (ecb_expect_false (maxreqs && !--maxreqs))
632 break;
633
634 if (maxtime)
635 {
636 gettimeofday (&tv_now, 0);
637
638 if (tvdiff (&tv_start, &tv_now) >= maxtime)
639 break;
640 }
641 }
642
643 errno = EAGAIN;
644 return -1;
645}
646
647static void
648etp_cancel (ETP_REQ *req)
649{
650 req->cancelled = 1;
651
652 eio_grp_cancel (req);
653}
654
655static void
656etp_submit (ETP_REQ *req)
657{
658 req->pri -= ETP_PRI_MIN;
659
660 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
661 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
662
663 if (ecb_expect_false (req->type == EIO_GROUP))
664 {
665 /* I hope this is worth it :/ */
666 X_LOCK (reqlock);
667 ++nreqs;
668 X_UNLOCK (reqlock);
669
670 X_LOCK (reslock);
671
672 ++npending;
673
674 if (!reqq_push (&res_queue, req) && want_poll_cb)
675 want_poll_cb ();
676
677 X_UNLOCK (reslock);
678 }
679 else
680 {
681 X_LOCK (reqlock);
682 ++nreqs;
683 ++nready;
684 reqq_push (&req_queue, req);
685 X_COND_SIGNAL (reqwait);
686 X_UNLOCK (reqlock);
687
688 etp_maybe_start_thread ();
689 }
690}
691
692static void ecb_cold
693etp_set_max_poll_time (double nseconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
696 max_poll_time = nseconds * EIO_TICKS;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
698}
699
700static void ecb_cold
701etp_set_max_poll_reqs (unsigned int maxreqs)
702{
703 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
704 max_poll_reqs = maxreqs;
705 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
706}
707
708static void ecb_cold
709etp_set_max_idle (unsigned int nthreads)
710{
711 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
712 max_idle = nthreads;
713 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
714}
715
716static void ecb_cold
717etp_set_idle_timeout (unsigned int seconds)
718{
719 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
720 idle_timeout = seconds;
721 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
722}
723
724static void ecb_cold
725etp_set_min_parallel (unsigned int nthreads)
726{
727 if (wanted < nthreads)
728 wanted = nthreads;
729}
730
731static void ecb_cold
732etp_set_max_parallel (unsigned int nthreads)
733{
734 if (wanted > nthreads)
735 wanted = nthreads;
736
737 while (started > wanted)
738 etp_end_thread ();
739}
740 401
741/*****************************************************************************/ 402/*****************************************************************************/
742 403
743static void 404static void
744grp_try_feed (eio_req *grp) 405grp_try_feed (eio_req *grp)
745{ 406{
746 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
747 { 408 {
748 grp->flags &= ~EIO_FLAG_GROUPADD; 409 grp->flags &= ~ETP_FLAG_GROUPADD;
749 410
750 EIO_FEED (grp); 411 EIO_FEED (grp);
751 412
752 /* stop if no progress has been made */ 413 /* stop if no progress has been made */
753 if (!(grp->flags & EIO_FLAG_GROUPADD)) 414 if (!(grp->flags & ETP_FLAG_GROUPADD))
754 { 415 {
755 grp->feed = 0; 416 grp->feed = 0;
756 break; 417 break;
757 } 418 }
758 } 419 }
765 426
766 /* call feeder, if applicable */ 427 /* call feeder, if applicable */
767 grp_try_feed (grp); 428 grp_try_feed (grp);
768 429
769 /* finish, if done */ 430 /* finish, if done */
770 if (!grp->size && grp->int1) 431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
771 return eio_finish (grp); 432 return eio_finish (grp);
772 else 433 else
773 return 0; 434 return 0;
774} 435}
775 436
811} 472}
812 473
813void 474void
814eio_grp_cancel (eio_req *grp) 475eio_grp_cancel (eio_req *grp)
815{ 476{
816 for (grp = grp->grp_first; grp; grp = grp->grp_next) 477 etp_grp_cancel (EIO_POOL, grp);
817 eio_cancel (grp);
818} 478}
819 479
820void 480void
821eio_cancel (eio_req *req) 481eio_cancel (eio_req *req)
822{ 482{
823 etp_cancel (req); 483 etp_cancel (EIO_POOL, req);
824} 484}
825 485
826void 486void
827eio_submit (eio_req *req) 487eio_submit (eio_req *req)
828{ 488{
829 etp_submit (req); 489 etp_submit (EIO_POOL, req);
830} 490}
831 491
832unsigned int 492unsigned int
833eio_nreqs (void) 493eio_nreqs (void)
834{ 494{
835 return etp_nreqs (); 495 return etp_nreqs (EIO_POOL);
836} 496}
837 497
838unsigned int 498unsigned int
839eio_nready (void) 499eio_nready (void)
840{ 500{
841 return etp_nready (); 501 return etp_nready (EIO_POOL);
842} 502}
843 503
844unsigned int 504unsigned int
845eio_npending (void) 505eio_npending (void)
846{ 506{
847 return etp_npending (); 507 return etp_npending (EIO_POOL);
848} 508}
849 509
850unsigned int ecb_cold 510unsigned int ecb_cold
851eio_nthreads (void) 511eio_nthreads (void)
852{ 512{
853 return etp_nthreads (); 513 return etp_nthreads (EIO_POOL);
854} 514}
855 515
856void ecb_cold 516void ecb_cold
857eio_set_max_poll_time (double nseconds) 517eio_set_max_poll_time (double nseconds)
858{ 518{
859 etp_set_max_poll_time (nseconds); 519 etp_set_max_poll_time (EIO_POOL, nseconds);
860} 520}
861 521
862void ecb_cold 522void ecb_cold
863eio_set_max_poll_reqs (unsigned int maxreqs) 523eio_set_max_poll_reqs (unsigned int maxreqs)
864{ 524{
865 etp_set_max_poll_reqs (maxreqs); 525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
866} 526}
867 527
868void ecb_cold 528void ecb_cold
869eio_set_max_idle (unsigned int nthreads) 529eio_set_max_idle (unsigned int nthreads)
870{ 530{
871 etp_set_max_idle (nthreads); 531 etp_set_max_idle (EIO_POOL, nthreads);
872} 532}
873 533
874void ecb_cold 534void ecb_cold
875eio_set_idle_timeout (unsigned int seconds) 535eio_set_idle_timeout (unsigned int seconds)
876{ 536{
877 etp_set_idle_timeout (seconds); 537 etp_set_idle_timeout (EIO_POOL, seconds);
878} 538}
879 539
880void ecb_cold 540void ecb_cold
881eio_set_min_parallel (unsigned int nthreads) 541eio_set_min_parallel (unsigned int nthreads)
882{ 542{
883 etp_set_min_parallel (nthreads); 543 etp_set_min_parallel (EIO_POOL, nthreads);
884} 544}
885 545
886void ecb_cold 546void ecb_cold
887eio_set_max_parallel (unsigned int nthreads) 547eio_set_max_parallel (unsigned int nthreads)
888{ 548{
889 etp_set_max_parallel (nthreads); 549 etp_set_max_parallel (EIO_POOL, nthreads);
890} 550}
891 551
892int eio_poll (void) 552int eio_poll (void)
893{ 553{
894 return etp_poll (); 554 return etp_poll (EIO_POOL);
895} 555}
896 556
897/*****************************************************************************/ 557/*****************************************************************************/
898/* work around various missing functions */ 558/* work around various missing functions */
899 559
900#if !HAVE_PREADWRITE 560#if HAVE_POSIX_CLOSE && !__linux
901# undef pread 561# define eio__close(fd) posix_close (fd, 0)
902# undef pwrite 562#else
903# define pread eio__pread 563# define eio__close(fd) close (fd)
904# define pwrite eio__pwrite
905
906static eio_ssize_t
907eio__pread (int fd, void *buf, size_t count, off_t offset)
908{
909 eio_ssize_t res;
910 off_t ooffset;
911
912 X_LOCK (preadwritelock);
913 ooffset = lseek (fd, 0, SEEK_CUR);
914 lseek (fd, offset, SEEK_SET);
915 res = read (fd, buf, count);
916 lseek (fd, ooffset, SEEK_SET);
917 X_UNLOCK (preadwritelock);
918
919 return res;
920}
921
922static eio_ssize_t
923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
924{
925 eio_ssize_t res;
926 off_t ooffset;
927
928 X_LOCK (preadwritelock);
929 ooffset = lseek (fd, 0, SEEK_CUR);
930 lseek (fd, offset, SEEK_SET);
931 res = write (fd, buf, count);
932 lseek (fd, ooffset, SEEK_SET);
933 X_UNLOCK (preadwritelock);
934
935 return res;
936}
937#endif 564#endif
565
566/* close() without disturbing errno */
567static void
568silent_close (int fd)
569{
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573}
938 574
939#ifndef HAVE_UTIMES 575#ifndef HAVE_UTIMES
940 576
941# undef utimes 577# undef utimes
942# define utimes(path,times) eio__utimes (path, times) 578# define utimes(path,times) eio__utimes (path, times)
984 int res; 620 int res;
985 621
986#if HAVE_SYS_SYNCFS 622#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd)); 623 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else 624#else
989 res = -1; 625 res = EIO_ENOSYS ();
990 errno = ENOSYS;
991#endif 626#endif
992 627
993 if (res < 0 && errno == ENOSYS && fd >= 0) 628 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync (); 629 sync ();
995 630
1025} 660}
1026 661
1027static int 662static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len) 663eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{ 664{
1030#if HAVE_FALLOCATE 665#if HAVE_LINUX_FALLOCATE
1031 return fallocate (fd, mode, offset, len); 666 return fallocate (fd, mode, offset, len);
1032#else 667#else
1033 errno = ENOSYS; 668 return EIO_ENOSYS ();
1034 return -1;
1035#endif 669#endif
1036} 670}
1037 671
1038#if !HAVE_READAHEAD 672#if !HAVE_READAHEAD
1039# undef readahead 673# undef readahead
1054 todo -= len; 688 todo -= len;
1055 } 689 }
1056 690
1057 FUBd; 691 FUBd;
1058 692
1059 errno = 0; 693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
1060 return count; 695 return 0;
1061} 696}
1062 697
1063#endif 698#endif
1064 699
1065/* sendfile always needs emulation */ 700/* sendfile always needs emulation */
1100 735
1101 /* according to source inspection, this is correct, and useful behaviour */ 736 /* according to source inspection, this is correct, and useful behaviour */
1102 if (sbytes) 737 if (sbytes)
1103 res = sbytes; 738 res = sbytes;
1104 739
1105# elif defined (__APPLE__) 740# elif defined __APPLE__
1106 off_t sbytes = count; 741 off_t sbytes = count;
1107 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1108 743
1109 /* according to the manpage, sbytes is always valid */ 744 /* according to the manpage, sbytes is always valid */
1110 if (sbytes) 745 if (sbytes)
1137 HANDLE h = TO_SOCKET (ifd); 772 HANDLE h = TO_SOCKET (ifd);
1138 SetFilePointer (h, offset, 0, FILE_BEGIN); 773 SetFilePointer (h, offset, 0, FILE_BEGIN);
1139 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1140 775
1141#else 776#else
1142 res = -1; 777 res = EIO_ENOSYS ();
1143 errno = ENOSYS;
1144#endif 778#endif
1145 779
1146 /* we assume sendfile can copy at least 128mb in one go */ 780 /* we assume sendfile can copy at least 128mb in one go */
1147 if (res <= 128 * 1024 * 1024) 781 if (res <= 128 * 1024 * 1024)
1148 { 782 {
1335 969
1336/*****************************************************************************/ 970/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */ 971/* requests implemented outside eio_execute, because they are so large */
1338 972
1339static void 973static void
1340eio__realpath (eio_req *req, etp_worker *self) 974eio__lseek (eio_req *req)
1341{ 975{
1342 char *rel = req->ptr1; 976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985}
986
987/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988static int
989eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990{
1343 char *res; 991 char *res;
992 const char *rel = path;
1344 char *tmp1, *tmp2; 993 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32 994#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX; 995 int symlinks = SYMLOOP_MAX;
1347#else 996#else
1348 int symlinks = 32; 997 int symlinks = 32;
1349#endif 998#endif
1350 999
1351 req->result = -1;
1352
1353 errno = EINVAL; 1000 errno = EINVAL;
1354 if (!rel) 1001 if (!rel)
1355 return; 1002 return -1;
1356 1003
1357 errno = ENOENT; 1004 errno = ENOENT;
1358 if (!*rel) 1005 if (!*rel)
1359 return; 1006 return -1;
1360 1007
1361 if (!req->ptr2) 1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1362 { 1009#ifdef _WIN32
1363 X_LOCK (wrklock); 1010 if (_access (rel, 4) != 0)
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return; 1011 return -1;
1371 }
1372 1012
1373 res = req->ptr2; 1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025#else
1374 tmp1 = res + PATH_MAX; 1026 tmp1 = res + EIO_PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX; 1027 tmp2 = tmp1 + EIO_PATH_MAX;
1376 1028
1377#if 0 /* disabled, the musl way to do things is just too racy */ 1029#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1030#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1379 /* on linux we may be able to ask the kernel */ 1031 /* on linux we may be able to ask the kernel */
1380 { 1032 {
1381 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); 1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1382 1034
1383 if (fd >= 0) 1035 if (fd >= 0)
1384 { 1036 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd); 1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX); 1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1387 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */ 1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1390 1041
1391 if (req->result > 0) 1042 if (req->result > 0)
1392 goto done; 1043 goto done;
1393 } 1044 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return; 1046 return -1;
1396 } 1047 }
1397#endif 1048#endif
1398#endif 1049#endif
1399 1050
1400 if (*rel != '/') 1051 if (*rel != '/')
1401 { 1052 {
1402 if (!getcwd (res, PATH_MAX)) 1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1403 return; 1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1404 1068
1405 if (res [1]) /* only use if not / */ 1069 if (res [1]) /* only use if not / */
1406 res += strlen (res); 1070 res += len;
1407 } 1071 }
1408 1072
1409 while (*rel) 1073 while (*rel)
1410 { 1074 {
1411 eio_ssize_t len, linklen; 1075 eio_ssize_t len, linklen;
1412 char *beg = rel; 1076 const char *beg = rel;
1413 1077
1414 while (*rel && *rel != '/') 1078 while (*rel && *rel != '/')
1415 ++rel; 1079 ++rel;
1416 1080
1417 len = rel - beg; 1081 len = rel - beg;
1429 1093
1430 if (beg [1] == '.' && len == 2) 1094 if (beg [1] == '.' && len == 2)
1431 { 1095 {
1432 /* .. - back up one component, if possible */ 1096 /* .. - back up one component, if possible */
1433 1097
1434 while (res != req->ptr2) 1098 while (res != tmpbuf->ptr)
1435 if (*--res == '/') 1099 if (*--res == '/')
1436 break; 1100 break;
1437 1101
1438 continue; 1102 continue;
1439 } 1103 }
1440 } 1104 }
1441 1105
1442 errno = ENAMETOOLONG; 1106 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1) 1107 if (res + 1 + len + 1 >= tmp1)
1444 return; 1108 return -1;
1445 1109
1446 /* copy one component */ 1110 /* copy one component */
1447 *res = '/'; 1111 *res = '/';
1448 memcpy (res + 1, beg, len); 1112 memcpy (res + 1, beg, len);
1449 1113
1450 /* zero-terminate, for readlink */ 1114 /* zero-terminate, for readlink */
1451 res [len + 1] = 0; 1115 res [len + 1] = 0;
1452 1116
1453 /* now check if it's a symlink */ 1117 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1455 1119
1456 if (linklen < 0) 1120 if (linklen < 0)
1457 { 1121 {
1458 if (errno != EINVAL) 1122 if (errno != EINVAL)
1459 return; 1123 return -1;
1460 1124
1461 /* it's a normal directory. hopefully */ 1125 /* it's a normal directory. hopefully */
1462 res += len + 1; 1126 res += len + 1;
1463 } 1127 }
1464 else 1128 else
1465 { 1129 {
1466 /* yay, it was a symlink - build new path in tmp2 */ 1130 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel); 1131 int rellen = strlen (rel);
1468 1132
1469 errno = ENAMETOOLONG; 1133 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX) 1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1471 return; 1135 return -1;
1472 1136
1473 errno = ELOOP; 1137 errno = ELOOP;
1474 if (!--symlinks) 1138 if (!--symlinks)
1475 return; 1139 return -1;
1476 1140
1477 if (*tmp1 == '/') 1141 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */ 1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1479 1143
1480 /* we need to be careful, as rel might point into tmp2 already */ 1144 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/'; 1146 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen); 1147 memcpy (tmp2, tmp1, linklen);
1485 rel = tmp2; 1149 rel = tmp2;
1486 } 1150 }
1487 } 1151 }
1488 1152
1489 /* special case for the lone root path */ 1153 /* special case for the lone root path */
1490 if (res == req->ptr2) 1154 if (res == tmpbuf->ptr)
1491 *res++ = '/'; 1155 *res++ = '/';
1492 1156
1493 req->result = res - (char *)req->ptr2; 1157 return res - (char *)tmpbuf->ptr;
1494 1158#endif
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1497} 1159}
1498 1160
1499static signed char 1161static signed char
1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1162eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1501{ 1163{
1689#ifdef _WIN32 1351#ifdef _WIN32
1690 { 1352 {
1691 int len = strlen ((const char *)req->ptr1); 1353 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH); 1354 char *path = malloc (MAX_PATH);
1693 const char *fmt; 1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1694 1357
1695 if (!len) 1358 if (!len)
1696 fmt = "./*"; 1359 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') 1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1698 fmt = "%s*"; 1361 fmt = "%s*";
1699 else 1362 else
1700 fmt = "%s/*"; 1363 fmt = "%s/*";
1701 1364
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); 1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1703 dirp = FindFirstFile (path, &entp); 1366 dirp = FindFirstFile (path, &entp);
1704 free (path); 1367 free (path);
1705 1368
1706 if (dirp == INVALID_HANDLE_VALUE) 1369 if (dirp == INVALID_HANDLE_VALUE)
1707 { 1370 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */ 1371 /* should steal _dosmaperr */
1711 switch (GetLastError ()) 1372 switch (GetLastError ())
1712 { 1373 {
1713 case ERROR_FILE_NOT_FOUND: 1374 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0; 1375 req->result = 0;
1715 break; 1376 break;
1716 1377
1717 case ERROR_INVALID_NAME: 1378 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND: 1379 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES: 1380 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT; 1381 errno = ENOENT;
1721 break; 1382 break;
1722 1383
1723 case ERROR_NOT_ENOUGH_MEMORY: 1384 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM; 1385 errno = ENOMEM;
1725 break; 1386 break;
1726 1387
1727 default: 1388 default:
1728 errno = EINVAL; 1389 errno = EINVAL;
1729 break; 1390 break;
1730 } 1391 }
1392
1393 return;
1731 } 1394 }
1732 } 1395 }
1733#else 1396#else
1397 #if HAVE_AT
1398 if (req->wd)
1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1734 dirp = opendir (req->ptr1); 1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1735#endif 1418#endif
1736 1419
1737 if (req->flags & EIO_FLAG_PTR1_FREE) 1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1); 1421 free (req->ptr1);
1739 1422
1740 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1742 req->ptr2 = names = malloc (namesalloc); 1425 req->ptr2 = names = malloc (namesalloc);
1743 1426
1744 if (dirp && names && (!flags || dents)) 1427 if (!names || (flags && !dents))
1428 return;
1429
1745 for (;;) 1430 for (;;)
1746 { 1431 {
1747 int done; 1432 int done;
1748 1433
1749#ifdef _WIN32 1434#ifdef _WIN32
1750 done = !dirp; 1435 done = !dirp;
1751#else 1436#else
1752 errno = 0; 1437 errno = 0;
1753 entp = readdir (dirp); 1438 entp = readdir (dirp);
1754 done = !entp; 1439 done = !entp;
1755#endif 1440#endif
1756 1441
1757 if (done) 1442 if (done)
1758 { 1443 {
1759#ifndef _WIN32 1444#ifndef _WIN32
1760 int old_errno = errno; 1445 int old_errno = errno;
1761 closedir (dirp); 1446 closedir (dirp);
1762 errno = old_errno; 1447 errno = old_errno;
1763 1448
1764 if (errno) 1449 if (errno)
1765 break; 1450 break;
1766#endif 1451#endif
1767 1452
1768 /* sort etc. */ 1453 /* sort etc. */
1769 req->int1 = flags; 1454 req->int1 = flags;
1770 req->result = dentoffs; 1455 req->result = dentoffs;
1771 1456
1772 if (flags & EIO_READDIR_STAT_ORDER) 1457 if (flags & EIO_READDIR_STAT_ORDER)
1773 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1774 else if (flags & EIO_READDIR_DIRS_FIRST) 1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1775 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1776 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1777 else 1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
1486 break;
1487 }
1488
1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
1491
1492 /* skip . and .. entries */
1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1494 {
1495 int len = D_NAMLEN (entp) + 1;
1496
1497 while (ecb_expect_false (namesoffs + len > namesalloc))
1498 {
1499 namesalloc *= 2;
1500 req->ptr2 = names = realloc (names, namesalloc);
1501
1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1778 { 1513 {
1779 /* in this case, all is known, and we just put dirs first and sort them */ 1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1518 break;
1519 }
1520
1780 eio_dirent *oth = dents + dentoffs; 1521 ent = dents + dentoffs;
1781 eio_dirent *dir = dents;
1782 1522
1783 /* now partition dirs to the front, and non-dirs to the back */ 1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1784 /* by walking from both sides and swapping if necessary */ 1524 ent->namelen = len - 1;
1785 while (oth > dir) 1525 ent->inode = D_INO (entp);
1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1786 { 1585 {
1787 if (dir->type == EIO_DT_DIR) 1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1788 ++dir; 1587 ent->score = 1;
1789 else if ((--oth)->type == EIO_DT_DIR) 1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1790 { 1589 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1791 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1792
1793 ++dir;
1794 }
1795 } 1590 }
1796 1591 else if (ent->type == EIO_DT_DIR)
1797 /* now sort the dirs only (dirs all have the same score) */ 1592 ent->score = 0;
1798 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1799 } 1593 }
1800
1801 break;
1802 }
1803
1804 /* now add the entry to our list(s) */
1805 name = D_NAME (entp);
1806
1807 /* skip . and .. entries */
1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1809 {
1810 int len = D_NAMLEN (entp) + 1;
1811
1812 while (ecb_expect_false (namesoffs + len > namesalloc))
1813 {
1814 namesalloc *= 2;
1815 req->ptr2 = names = realloc (names, namesalloc);
1816
1817 if (!names)
1818 break;
1819 } 1594 }
1820 1595
1821 memcpy (names + namesoffs, name, len);
1822
1823 if (dents)
1824 {
1825 struct eio_dirent *ent;
1826
1827 if (ecb_expect_false (dentoffs == dentalloc))
1828 {
1829 dentalloc *= 2;
1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1831
1832 if (!dents)
1833 break;
1834 }
1835
1836 ent = dents + dentoffs;
1837
1838 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1839 ent->namelen = len - 1;
1840 ent->inode = D_INO (entp);
1841
1842 inode_bits |= ent->inode;
1843
1844 switch (D_TYPE (entp))
1845 {
1846 default:
1847 ent->type = EIO_DT_UNKNOWN;
1848 flags |= EIO_READDIR_FOUND_UNKNOWN;
1849 break;
1850
1851 #ifdef DT_FIFO
1852 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1853 #endif
1854 #ifdef DT_CHR
1855 case DT_CHR: ent->type = EIO_DT_CHR; break;
1856 #endif
1857 #ifdef DT_MPC
1858 case DT_MPC: ent->type = EIO_DT_MPC; break;
1859 #endif
1860 #ifdef DT_DIR
1861 case DT_DIR: ent->type = EIO_DT_DIR; break;
1862 #endif
1863 #ifdef DT_NAM
1864 case DT_NAM: ent->type = EIO_DT_NAM; break;
1865 #endif
1866 #ifdef DT_BLK
1867 case DT_BLK: ent->type = EIO_DT_BLK; break;
1868 #endif
1869 #ifdef DT_MPB
1870 case DT_MPB: ent->type = EIO_DT_MPB; break;
1871 #endif
1872 #ifdef DT_REG
1873 case DT_REG: ent->type = EIO_DT_REG; break;
1874 #endif
1875 #ifdef DT_NWK
1876 case DT_NWK: ent->type = EIO_DT_NWK; break;
1877 #endif
1878 #ifdef DT_CMP
1879 case DT_CMP: ent->type = EIO_DT_CMP; break;
1880 #endif
1881 #ifdef DT_LNK
1882 case DT_LNK: ent->type = EIO_DT_LNK; break;
1883 #endif
1884 #ifdef DT_SOCK
1885 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1886 #endif
1887 #ifdef DT_DOOR
1888 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1889 #endif
1890 #ifdef DT_WHT
1891 case DT_WHT: ent->type = EIO_DT_WHT; break;
1892 #endif
1893 }
1894
1895 ent->score = 7;
1896
1897 if (flags & EIO_READDIR_DIRS_FIRST)
1898 {
1899 if (ent->type == EIO_DT_UNKNOWN)
1900 {
1901 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1902 ent->score = 1;
1903 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1904 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1905 }
1906 else if (ent->type == EIO_DT_DIR)
1907 ent->score = 0;
1908 }
1909 }
1910
1911 namesoffs += len; 1596 namesoffs += len;
1912 ++dentoffs; 1597 ++dentoffs;
1913 } 1598 }
1914 1599
1915 if (EIO_CANCELLED (req)) 1600 if (EIO_CANCELLED (req))
1916 { 1601 {
1917 errno = ECANCELED; 1602 errno = ECANCELED;
1918 break; 1603 break;
1919 } 1604 }
1920 1605
1921#ifdef _WIN32 1606#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp)) 1607 if (!FindNextFile (dirp, &entp))
1923 { 1608 {
1924 FindClose (dirp); 1609 FindClose (dirp);
1925 dirp = 0; 1610 dirp = 0;
1926 } 1611 }
1927#endif 1612#endif
1928 } 1613 }
1929} 1614}
1615
1616/*****************************************************************************/
1617/* working directory stuff */
1618/* various deficiencies in the posix 2008 api force us to */
1619/* keep the absolute path in string form at all times */
1620/* fuck yeah. */
1621
1622#if !HAVE_AT
1623
1624/* a bit like realpath, but usually faster because it doesn'T have to return */
1625/* an absolute or canonical path */
1626static const char *
1627wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628{
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1635 {
1636 int l1 = wd->len;
1637 int l2 = strlen (path);
1638
1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1640
1641 memcpy (res, wd->str, l1);
1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647}
1648
1649#endif
1650
1651static eio_wd
1652eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1653{
1654 int fd;
1655 eio_wd res;
1656 int len = eio__realpath (tmpbuf, wd, path);
1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661#if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666#endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670#if HAVE_AT
1671 res->fd = fd;
1672#endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679}
1680
1681eio_wd
1682eio_wd_open_sync (eio_wd wd, const char *path)
1683{
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689}
1690
1691void
1692eio_wd_close_sync (eio_wd wd)
1693{
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701}
1702
1703#if HAVE_AT
1704
1705static int
1706eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707{
1708#if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1710#else
1711 if (flags)
1712 return EIO_ENOSYS ();
1713
1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1715#endif
1716}
1717
1718/* they forgot these */
1719
1720static int
1721eio__truncateat (int dirfd, const char *path, off_t length)
1722{
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1731 return res;
1732}
1733
1734static int
1735eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736{
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746}
1747
1748#endif
1930 1749
1931/*****************************************************************************/ 1750/*****************************************************************************/
1932 1751
1933#define ALLOC(len) \ 1752#define ALLOC(len) \
1934 if (!req->ptr2) \ 1753 if (!req->ptr2) \
1935 { \ 1754 { \
1936 X_LOCK (wrklock); \ 1755 X_LOCK (EIO_POOL->wrklock); \
1937 req->flags |= EIO_FLAG_PTR2_FREE; \ 1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1938 X_UNLOCK (wrklock); \ 1757 X_UNLOCK (EIO_POOL->wrklock); \
1939 req->ptr2 = malloc (len); \ 1758 req->ptr2 = malloc (len); \
1940 if (!req->ptr2) \ 1759 if (!req->ptr2) \
1941 { \ 1760 { \
1942 errno = ENOMEM; \ 1761 errno = ENOMEM; \
1943 req->result = -1; \ 1762 req->result = -1; \
1944 break; \ 1763 goto alloc_fail; \
1945 } \ 1764 } \
1946 } 1765 }
1947 1766
1948X_THREAD_PROC (etp_proc) 1767/*****************************************************************************/
1949{
1950 ETP_REQ *req;
1951 struct timespec ts;
1952 etp_worker *self = (etp_worker *)thr_arg;
1953 1768
1954#if HAVE_PRCTL_SET_NAME 1769static void
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0); 1770eio__slurp (int fd, eio_req *req)
1956#endif 1771{
1772 req->result = fd;
1957 1773
1958 /* try to distribute timeouts somewhat evenly */ 1774 if (fd < 0)
1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1775 return;
1960 1776
1961 for (;;) 1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1962 { 1778 {
1963 ts.tv_sec = 0; 1779 off_t size = lseek (fd, 0, SEEK_END);
1964 1780
1965 X_LOCK (reqlock);
1966
1967 for (;;)
1968 {
1969 self->req = req = reqq_shift (&req_queue);
1970
1971 if (req)
1972 break;
1973
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1975 {
1976 X_UNLOCK (reqlock);
1977 X_LOCK (wrklock);
1978 --started;
1979 X_UNLOCK (wrklock);
1980 goto quit;
1981 }
1982
1983 ++idle;
1984
1985 if (idle <= max_idle)
1986 /* we are allowed to idle, so do so without any timeout */
1987 X_COND_WAIT (reqwait, reqlock);
1988 else
1989 {
1990 /* initialise timeout once */
1991 if (!ts.tv_sec)
1992 ts.tv_sec = time (0) + idle_timeout;
1993
1994 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1995 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1996 }
1997
1998 --idle;
1999 }
2000
2001 --nready;
2002
2003 X_UNLOCK (reqlock);
2004
2005 if (req->type < 0) 1781 if (req->offs < 0)
2006 goto quit; 1782 req->offs += size;
2007 1783
2008 ETP_EXECUTE (self, req); 1784 if (!req->size)
2009 1785 req->size = size - req->offs;
2010 X_LOCK (reslock);
2011
2012 ++npending;
2013
2014 if (!reqq_push (&res_queue, req) && want_poll_cb)
2015 want_poll_cb ();
2016
2017 self->req = 0;
2018 etp_worker_clear (self);
2019
2020 X_UNLOCK (reslock);
2021 } 1786 }
2022 1787
2023quit: 1788 ALLOC (req->size);
2024 X_LOCK (wrklock); 1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
2025 etp_worker_free (self);
2026 X_UNLOCK (wrklock);
2027 1790
2028 return 0; 1791 silent_close (fd);
2029}
2030 1792
2031/*****************************************************************************/ 1793alloc_fail:
1794 ;
1795}
2032 1796
2033int ecb_cold 1797int ecb_cold
2034eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1798eio_init (void (*want_poll)(void), void (*done_poll)(void))
2035{ 1799{
2036#if !HAVE_PREADWRITE 1800 eio_want_poll_cb = want_poll;
2037 X_MUTEX_CREATE (preadwritelock); 1801 eio_done_poll_cb = done_poll;
2038#endif
2039 1802
2040 return etp_init (want_poll, done_poll); 1803 return etp_init (EIO_POOL, 0, 0, 0);
2041} 1804}
2042 1805
2043ecb_inline void 1806ecb_inline void
2044eio_api_destroy (eio_req *req) 1807eio_api_destroy (eio_req *req)
2045{ 1808{
2046 free (req); 1809 free (req);
2047} 1810}
2048 1811
2049#define REQ(rtype) \ 1812#define REQ(rtype) \
2050 eio_req *req; \ 1813 eio_req *req; \
2051 \ 1814 \
2052 req = (eio_req *)calloc (1, sizeof *req); \ 1815 req = (eio_req *)calloc (1, sizeof *req); \
2053 if (!req) \ 1816 if (!req) \
2054 return 0; \ 1817 return 0; \
2068 { \ 1831 { \
2069 eio_api_destroy (req); \ 1832 eio_api_destroy (req); \
2070 return 0; \ 1833 return 0; \
2071 } 1834 }
2072 1835
1836#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
2073static void 1838static void
2074eio_execute (etp_worker *self, eio_req *req) 1839eio_execute (etp_worker *self, eio_req *req)
2075{ 1840{
1841#if HAVE_AT
1842 int dirfd;
1843#else
1844 const char *path;
1845#endif
1846
2076 if (ecb_expect_false (EIO_CANCELLED (req))) 1847 if (ecb_expect_false (EIO_CANCELLED (req)))
2077 { 1848 {
2078 req->result = -1; 1849 req->result = -1;
2079 req->errorno = ECANCELED; 1850 req->errorno = ECANCELED;
2080 return; 1851 return;
2081 } 1852 }
2082 1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1869
2083 switch (req->type) 1870 switch (req->type)
2084 { 1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
2085 case EIO_READ: ALLOC (req->size); 1879 case EIO_READ: ALLOC (req->size);
2086 req->result = req->offs >= 0 1880 req->result = req->offs >= 0
2087 ? pread (req->int1, req->ptr2, req->size, req->offs) 1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
2088 : read (req->int1, req->ptr2, req->size); break; 1882 : read (req->int1, req->ptr2, req->size); break;
2089 case EIO_WRITE: req->result = req->offs >= 0 1883 case EIO_WRITE: req->result = req->offs >= 0
2090 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2091 : write (req->int1, req->ptr2, req->size); break; 1885 : write (req->int1, req->ptr2, req->size); break;
2092 1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
2093 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; 1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2095 1892
1893#if HAVE_AT
1894
2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1899 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1900 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1901 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1902 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1903 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1904
1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1921 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1927 if (req->result == EIO_PATH_MAX)
1928 {
1929 req->result = -1;
1930 errno = ENAMETOOLONG;
1931 }
1932 break;
1933 case EIO_UTIME:
1934 case EIO_FUTIME:
1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1957#else
1958
1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1968
1969 case EIO_UNLINK: req->result = unlink (path ); break;
1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1986
1987 case EIO_UTIME:
1988 case EIO_FUTIME:
1989 {
1990 struct timeval tv[2];
1991 struct timeval *times;
1992
1993 if (req->nv1 != -1. || req->nv2 != -1.)
1994 {
1995 tv[0].tv_sec = req->nv1;
1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1997 tv[1].tv_sec = req->nv2;
1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1999
2000 times = tv;
2001 }
2002 else
2003 times = 0;
2004
2005 req->result = req->type == EIO_FUTIME
2006 ? futimes (req->int1, times)
2007 : utimes (req->ptr1, times);
2008 }
2009 break;
2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2102 2022
2103 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2104 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2105 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2106 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2107 2025
2108 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2109 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2110 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2111 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2112 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2113 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2114 2029
2115 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2116 case EIO_CLOSE: req->result = close (req->int1); break; 2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2117 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2118 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2119 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2120 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
2128 case EIO_READLINK: ALLOC (PATH_MAX);
2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2130
2131 case EIO_SYNC: req->result = 0; sync (); break; 2032 case EIO_SYNC: req->result = 0; sync (); break;
2132 case EIO_FSYNC: req->result = fsync (req->int1); break; 2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2134 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; 2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2154 req->result = select (0, 0, 0, 0, &tv); 2055 req->result = select (0, 0, 0, 0, &tv);
2155 } 2056 }
2156#endif 2057#endif
2157 break; 2058 break;
2158 2059
2159 case EIO_UTIME: 2060#if 0
2160 case EIO_FUTIME:
2161 {
2162 struct timeval tv[2];
2163 struct timeval *times;
2164
2165 if (req->nv1 != -1. || req->nv2 != -1.)
2166 {
2167 tv[0].tv_sec = req->nv1;
2168 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2169 tv[1].tv_sec = req->nv2;
2170 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2171
2172 times = tv;
2173 }
2174 else
2175 times = 0;
2176
2177 req->result = req->type == EIO_FUTIME
2178 ? futimes (req->int1, times)
2179 : utimes (req->ptr1, times);
2180 }
2181 break;
2182
2183 case EIO_GROUP: 2061 case EIO_GROUP:
2184 abort (); /* handled in eio_request */ 2062 abort (); /* handled in eio_request */
2063#endif
2185 2064
2186 case EIO_NOP: 2065 case EIO_NOP:
2187 req->result = 0; 2066 req->result = 0;
2188 break; 2067 break;
2189 2068
2190 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
2191 req->feed (req); 2070 req->feed (req);
2192 break; 2071 break;
2193 2072
2194 default: 2073 default:
2195 errno = ENOSYS;
2196 req->result = -1; 2074 req->result = EIO_ENOSYS ();
2197 break; 2075 break;
2198 } 2076 }
2199 2077
2078alloc_fail:
2200 req->errorno = errno; 2079 req->errorno = errno;
2201} 2080}
2202 2081
2203#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
2204 2083
2084eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087}
2088
2089eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092}
2093
2205eio_req *eio_nop (int pri, eio_cb cb, void *data) 2094eio_req *eio_nop (int pri, eio_cb cb, void *data)
2206{ 2095{
2207 REQ (EIO_NOP); SEND; 2096 REQ (EIO_NOP); SEND;
2208} 2097}
2209 2098
2270eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2159eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2271{ 2160{
2272 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2273} 2162}
2274 2163
2164eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167}
2168
2275eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2169eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2276{ 2170{
2277 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2278} 2172}
2279 2173
2280eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2174eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2281{ 2175{
2282 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2177}
2178
2179eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182}
2183
2184eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2283} 2187}
2284 2188
2285eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2286{ 2190{
2287 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2432eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2433{ 2337{
2434 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2435} 2339}
2436 2340
2341eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342{
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344}
2345
2437eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) 2346eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2438{ 2347{
2439 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2440} 2349}
2441 2350
2475void 2384void
2476eio_grp_add (eio_req *grp, eio_req *req) 2385eio_grp_add (eio_req *grp, eio_req *req)
2477{ 2386{
2478 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2479 2388
2480 grp->flags |= EIO_FLAG_GROUPADD; 2389 grp->flags |= ETP_FLAG_GROUPADD;
2481 2390
2482 ++grp->size; 2391 ++grp->size;
2483 req->grp = grp; 2392 req->grp = grp;
2484 2393
2485 req->grp_prev = 0; 2394 req->grp_prev = 0;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines