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

Comparing libeio/eio.c (file contents):
Revision 1.83 by root, Thu Jul 7 22:36:18 2011 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 2012 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 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 *
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>
54#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h> 55#include <string.h>
56#include <errno.h> 56#include <errno.h>
57#include <sys/types.h> 57#include <sys/types.h>
58#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 59#include <limits.h>
61#include <fcntl.h> 60#include <fcntl.h>
62#include <assert.h> 61#include <assert.h>
63 62
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
68#endif 67#endif
69 68
70#ifndef ECANCELED 69#ifndef ECANCELED
71# define ECANCELED EDOM 70# define ECANCELED EDOM
72#endif 71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
73 79
74static void eio_destroy (eio_req *req); 80static void eio_destroy (eio_req *req);
75 81
76#ifndef EIO_FINISH 82#ifndef EIO_FINISH
77# 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
83 89
84#ifndef EIO_FEED 90#ifndef EIO_FEED
85# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
86#endif 92#endif
87 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
88#ifdef _WIN32 105#ifdef _WIN32
89 106
90 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
91#else 197#else
92 198
93# include <sys/time.h> 199 #include <sys/time.h>
94# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
95# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
96# include <utime.h> 237# include <utime.h>
97# include <signal.h>
98# include <dirent.h>
99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif 238#endif
112 239
113#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 241# include <sys/syscall.h>
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif 242#endif
121 243
122# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
123# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
124# endif
125
126#endif 246#endif
127 247
128#if HAVE_SENDFILE 248#if HAVE_SENDFILE
129# if __linux 249# if __linux
130# include <sys/sendfile.h> 250# include <sys/sendfile.h>
145#endif 265#endif
146#ifndef D_INO 266#ifndef D_INO
147# define D_INO(de) 0 267# define D_INO(de) 0
148#endif 268#endif
149#ifndef D_NAMLEN 269#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 271#endif
152 272
153/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 274#ifndef NAME_MAX
155# define NAME_MAX 4096 275# define NAME_MAX 4096
162 282
163/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
165 285
166#define dBUF \ 286#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 288 errno = ENOMEM; \
172 if (!eio_buf) \ 289 if (!eio_buf) \
173 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
174 294
175#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
176 340
177#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
179 343
180struct etp_worker; 344struct etp_worker;
184static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 352
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/ 353/*****************************************************************************/
207 354
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 356
210/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
221static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
222 369
223static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
225 372
226static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231 378
232static xmutex_t wrklock; 379static xmutex_t wrklock;
233static xmutex_t reslock; 380static xmutex_t reslock;
234static xmutex_t reqlock; 381static xmutex_t reqlock;
235static xcond_t reqwait; 382static xcond_t reqwait;
238/* 385/*
239 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot, 387 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine. 388 * but that's your problem, not mine.
242 */ 389 */
243static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
244#endif 391#endif
245 392
246typedef struct etp_worker 393typedef struct etp_worker
247{ 394{
395 struct tmpbuf tmpbuf;
396
248 /* locked by wrklock */ 397 /* locked by wrklock */
249 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
250 399
251 xthread_t tid; 400 xthread_t tid;
252 401
253 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
257} etp_worker; 405} etp_worker;
258 406
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
260 408
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263 411
264/* worker threads management */ 412/* worker threads management */
265 413
266static void ecb_cold 414static void
267etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
268{ 416{
269 ETP_WORKER_CLEAR (wrk);
270} 417}
271 418
272static void ecb_cold 419static void ecb_cold
273etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
274{ 421{
422 free (wrk->tmpbuf.ptr);
423
275 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
277 426
278 free (wrk); 427 free (wrk);
279} 428}
335} etp_reqq; 484} etp_reqq;
336 485
337static etp_reqq req_queue; 486static etp_reqq req_queue;
338static etp_reqq res_queue; 487static etp_reqq res_queue;
339 488
489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
340static int ecb_noinline 500static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
342{ 502{
343 int pri = req->pri; 503 int pri = req->pri;
344 req->next = 0; 504 req->next = 0;
378 } 538 }
379 539
380 abort (); 540 abort ();
381} 541}
382 542
383static void ecb_cold 543static int ecb_cold
384etp_thread_init (void) 544etp_init (void (*want_poll)(void), void (*done_poll)(void))
385{ 545{
386 X_MUTEX_CREATE (wrklock); 546 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock); 547 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock); 548 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait); 549 X_COND_CREATE (reqwait);
390}
391 550
392static void ecb_cold 551 reqq_init (&req_queue);
393etp_atfork_prepare (void) 552 reqq_init (&res_queue);
394{
395 X_LOCK (wrklock);
396 X_LOCK (reqlock);
397 X_LOCK (reslock);
398#if !HAVE_PREADWRITE
399 X_LOCK (preadwritelock);
400#endif
401}
402 553
403static void ecb_cold 554 wrk_first.next =
404etp_atfork_parent (void) 555 wrk_first.prev = &wrk_first;
405{
406#if !HAVE_PREADWRITE
407 X_UNLOCK (preadwritelock);
408#endif
409 X_UNLOCK (reslock);
410 X_UNLOCK (reqlock);
411 X_UNLOCK (wrklock);
412}
413
414static void ecb_cold
415etp_atfork_child (void)
416{
417 ETP_REQ *prv;
418
419 while ((prv = reqq_shift (&req_queue)))
420 ETP_DESTROY (prv);
421
422 while ((prv = reqq_shift (&res_queue)))
423 ETP_DESTROY (prv);
424
425 while (wrk_first.next != &wrk_first)
426 {
427 etp_worker *wrk = wrk_first.next;
428
429 if (wrk->req)
430 ETP_DESTROY (wrk->req);
431
432 etp_worker_clear (wrk);
433 etp_worker_free (wrk);
434 }
435 556
436 started = 0; 557 started = 0;
437 idle = 0; 558 idle = 0;
438 nreqs = 0; 559 nreqs = 0;
439 nready = 0; 560 nready = 0;
440 npending = 0; 561 npending = 0;
441
442 etp_thread_init ();
443}
444
445static void ecb_cold
446etp_once_init (void)
447{
448 etp_thread_init ();
449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450}
451
452static int ecb_cold
453etp_init (void (*want_poll)(void), void (*done_poll)(void))
454{
455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457 pthread_once (&doinit, etp_once_init);
458 562
459 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
460 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
461 565
462 return 0; 566 return 0;
502} 606}
503 607
504static void ecb_cold 608static void ecb_cold
505etp_end_thread (void) 609etp_end_thread (void)
506{ 610{
507 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
508 612
509 req->type = -1; 613 req->type = -1;
510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
511 615
512 X_LOCK (reqlock); 616 X_LOCK (reqlock);
845# undef pread 949# undef pread
846# undef pwrite 950# undef pwrite
847# define pread eio__pread 951# define pread eio__pread
848# define pwrite eio__pwrite 952# define pwrite eio__pwrite
849 953
850static ssize_t 954static eio_ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
852{ 956{
853 ssize_t res; 957 eio_ssize_t res;
854 off_t ooffset; 958 off_t ooffset;
855 959
856 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
861 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
862 966
863 return res; 967 return res;
864} 968}
865 969
866static ssize_t 970static eio_ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{ 972{
869 ssize_t res; 973 eio_ssize_t res;
870 off_t ooffset; 974 off_t ooffset;
871 975
872 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
919 1023
920#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
921# undef fdatasync 1025# undef fdatasync
922# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
923#endif 1027#endif
1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = EIO_ENOSYS ();
1038#endif
1039
1040 if (res < 0 && errno == ENOSYS && fd >= 0)
1041 sync ();
1042
1043 return res;
1044}
924 1045
925/* sync_file_range always needs emulation */ 1046/* sync_file_range always needs emulation */
926static int 1047static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1048eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{ 1049{
949 * call fdatasync, as that matches the expectation of its users best */ 1070 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 1071 return fdatasync (fd);
951} 1072}
952 1073
953static int 1074static int
954eio__fallocate (int fd, int mode, off_t offset, size_t nbytes) 1075eio__fallocate (int fd, int mode, off_t offset, size_t len)
955{ 1076{
956#if HAVE_FALLOCATE 1077#if HAVE_LINUX_FALLOCATE
957 return fallocate (fd, offset, nbytes, flags); 1078 return fallocate (fd, mode, offset, len);
958#else 1079#else
959 errno = ENOSYS; 1080 return EIO_ENOSYS ();
960 return -1;
961#endif 1081#endif
962} 1082}
963 1083
964#if !HAVE_READAHEAD 1084#if !HAVE_READAHEAD
965# undef readahead 1085# undef readahead
966# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1086# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
967 1087
968static ssize_t 1088static eio_ssize_t
969eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1089eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
970{ 1090{
971 size_t todo = count; 1091 size_t todo = count;
972 dBUF; 1092 dBUF;
973 1093
978 pread (fd, eio_buf, len, offset); 1098 pread (fd, eio_buf, len, offset);
979 offset += len; 1099 offset += len;
980 todo -= len; 1100 todo -= len;
981 } 1101 }
982 1102
983 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
984 return count; 1107 return 0;
985} 1108}
986 1109
987#endif 1110#endif
988 1111
989/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
990static ssize_t 1113static eio_ssize_t
991eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
992{ 1115{
993 ssize_t written = 0; 1116 eio_ssize_t written = 0;
994 ssize_t res; 1117 eio_ssize_t res;
995 1118
996 if (!count) 1119 if (!count)
997 return 0; 1120 return 0;
998 1121
999 for (;;) 1122 for (;;)
1024 1147
1025 /* according to source inspection, this is correct, and useful behaviour */ 1148 /* according to source inspection, this is correct, and useful behaviour */
1026 if (sbytes) 1149 if (sbytes)
1027 res = sbytes; 1150 res = sbytes;
1028 1151
1029# elif defined (__APPLE__) 1152# elif defined __APPLE__
1030 off_t sbytes = count; 1153 off_t sbytes = count;
1031 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1032 1155
1033 /* according to the manpage, sbytes is always valid */ 1156 /* according to the manpage, sbytes is always valid */
1034 if (sbytes) 1157 if (sbytes)
1051 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
1052 res = sbytes; 1175 res = sbytes;
1053 1176
1054# endif 1177# endif
1055 1178
1056#elif defined (_WIN32) 1179#elif defined (_WIN32) && 0
1057 /* does not work, just for documentation of what would need to be done */ 1180 /* does not work, just for documentation of what would need to be done */
1058 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1059 /* libeio guarantees that the file offset does not change, and windows */ 1182 /* libeio guarantees that the file offset does not change, and windows */
1060 /* has no way to get an independent handle to the same file description */ 1183 /* has no way to get an independent handle to the same file description */
1061 HANDLE h = TO_SOCKET (ifd); 1184 HANDLE h = TO_SOCKET (ifd);
1062 SetFilePointer (h, offset, 0, FILE_BEGIN); 1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1063 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1064 1187
1065#else 1188#else
1066 res = -1; 1189 res = EIO_ENOSYS ();
1067 errno = ENOSYS;
1068#endif 1190#endif
1069 1191
1070 /* we assume sendfile can copy at least 128mb in one go */ 1192 /* we assume sendfile can copy at least 128mb in one go */
1071 if (res <= 128 * 1024 * 1024) 1193 if (res <= 128 * 1024 * 1024)
1072 { 1194 {
1094 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1216 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1095 /* BSDs */ 1217 /* BSDs */
1096#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1218#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1097 || errno == ENOTSUP 1219 || errno == ENOTSUP
1098#endif 1220#endif
1221#ifdef EOPNOTSUPP /* windows */
1099 || errno == EOPNOTSUPP /* BSDs */ 1222 || errno == EOPNOTSUPP /* BSDs */
1223#endif
1100#if __solaris 1224#if __solaris
1101 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1225 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1102#endif 1226#endif
1103 ) 1227 )
1104 ) 1228 )
1108 1232
1109 res = 0; 1233 res = 0;
1110 1234
1111 while (count) 1235 while (count)
1112 { 1236 {
1113 ssize_t cnt; 1237 eio_ssize_t cnt;
1114 1238
1115 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1239 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1116 1240
1117 if (cnt <= 0) 1241 if (cnt <= 0)
1118 { 1242 {
1130 1254
1131 offset += cnt; 1255 offset += cnt;
1132 res += cnt; 1256 res += cnt;
1133 count -= cnt; 1257 count -= cnt;
1134 } 1258 }
1259
1260 FUBd;
1135 } 1261 }
1136 1262
1137 return res; 1263 return res;
1138} 1264}
1139 1265
1166 /* round up length */ 1292 /* round up length */
1167 *length = (*length + mask) & ~mask; 1293 *length = (*length + mask) & ~mask;
1168} 1294}
1169 1295
1170#if !_POSIX_MEMLOCK 1296#if !_POSIX_MEMLOCK
1171# define eio__mlockall(a) ((errno = ENOSYS), -1) 1297# define eio__mlockall(a) EIO_ENOSYS ()
1172#else 1298#else
1173 1299
1174static int 1300static int
1175eio__mlockall (int flags) 1301eio__mlockall (int flags)
1176{ 1302{
1190 return mlockall (flags); 1316 return mlockall (flags);
1191} 1317}
1192#endif 1318#endif
1193 1319
1194#if !_POSIX_MEMLOCK_RANGE 1320#if !_POSIX_MEMLOCK_RANGE
1195# define eio__mlock(a,b) ((errno = ENOSYS), -1) 1321# define eio__mlock(a,b) EIO_ENOSYS ()
1196#else 1322#else
1197 1323
1198static int 1324static int
1199eio__mlock (void *addr, size_t length) 1325eio__mlock (void *addr, size_t length)
1200{ 1326{
1204} 1330}
1205 1331
1206#endif 1332#endif
1207 1333
1208#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1334#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1209# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1335# define eio__msync(a,b,c) EIO_ENOSYS ()
1210#else 1336#else
1211 1337
1212static int 1338static int
1213eio__msync (void *mem, size_t len, int flags) 1339eio__msync (void *mem, size_t len, int flags)
1214{ 1340{
1255 1381
1256/*****************************************************************************/ 1382/*****************************************************************************/
1257/* requests implemented outside eio_execute, because they are so large */ 1383/* requests implemented outside eio_execute, because they are so large */
1258 1384
1259static void 1385static void
1260eio__realpath (eio_req *req, etp_worker *self) 1386eio__lseek (eio_req *req)
1261{ 1387{
1262 char *rel = req->ptr1; 1388 /* this usually gets optimised away completely, or your compiler sucks, */
1389 /* or the whence constants really are not 0, 1, 2 */
1390 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1391 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1392 : req->int2 == EIO_SEEK_END ? SEEK_END
1393 : req->int2;
1394
1395 req->offs = lseek (req->int1, req->offs, whence);
1396 req->result = req->offs == (off_t)-1 ? -1 : 0;
1397}
1398
1399/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1400static int
1401eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1402{
1403 const char *rel = path;
1263 char *res; 1404 char *res;
1264 char *tmp1, *tmp2; 1405 char *tmp1, *tmp2;
1265#if SYMLOOP_MAX > 32 1406#if SYMLOOP_MAX > 32
1266 int symlinks = SYMLOOP_MAX; 1407 int symlinks = SYMLOOP_MAX;
1267#else 1408#else
1268 int symlinks = 32; 1409 int symlinks = 32;
1269#endif 1410#endif
1270 1411
1271 req->result = -1;
1272
1273 errno = EINVAL; 1412 errno = EINVAL;
1274 if (!rel) 1413 if (!rel)
1275 return; 1414 return -1;
1276 1415
1277 errno = ENOENT; 1416 errno = ENOENT;
1278 if (!*rel) 1417 if (!*rel)
1279 return; 1418 return -1;
1280 1419
1281 if (!req->ptr2) 1420 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1282 {
1283 X_LOCK (wrklock);
1284 req->flags |= EIO_FLAG_PTR2_FREE;
1285 X_UNLOCK (wrklock);
1286 req->ptr2 = malloc (PATH_MAX * 3);
1287
1288 errno = ENOMEM;
1289 if (!req->ptr2)
1290 return;
1291 }
1292
1293 res = req->ptr2;
1294 tmp1 = res + PATH_MAX; 1421 tmp1 = res + PATH_MAX;
1295 tmp2 = tmp1 + PATH_MAX; 1422 tmp2 = tmp1 + PATH_MAX;
1296 1423
1297#if 0 /* disabled, the musl way to do things is just too racy */ 1424#if 0 /* disabled, the musl way to do things is just too racy */
1298#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1425#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1317#endif 1444#endif
1318#endif 1445#endif
1319 1446
1320 if (*rel != '/') 1447 if (*rel != '/')
1321 { 1448 {
1449 int len;
1450
1451 errno = ENOENT;
1452 if (wd == EIO_INVALID_WD)
1453 return -1;
1454
1455 if (wd == EIO_CWD)
1456 {
1322 if (!getcwd (res, PATH_MAX)) 1457 if (!getcwd (res, PATH_MAX))
1323 return; 1458 return -1;
1459
1460 len = strlen (res);
1461 }
1462 else
1463 memcpy (res, wd->str, len = wd->len);
1324 1464
1325 if (res [1]) /* only use if not / */ 1465 if (res [1]) /* only use if not / */
1326 res += strlen (res); 1466 res += len;
1327 } 1467 }
1328 1468
1329 while (*rel) 1469 while (*rel)
1330 { 1470 {
1331 ssize_t len, linklen; 1471 eio_ssize_t len, linklen;
1332 char *beg = rel; 1472 const char *beg = rel;
1333 1473
1334 while (*rel && *rel != '/') 1474 while (*rel && *rel != '/')
1335 ++rel; 1475 ++rel;
1336 1476
1337 len = rel - beg; 1477 len = rel - beg;
1349 1489
1350 if (beg [1] == '.' && len == 2) 1490 if (beg [1] == '.' && len == 2)
1351 { 1491 {
1352 /* .. - back up one component, if possible */ 1492 /* .. - back up one component, if possible */
1353 1493
1354 while (res != req->ptr2) 1494 while (res != tmpbuf->ptr)
1355 if (*--res == '/') 1495 if (*--res == '/')
1356 break; 1496 break;
1357 1497
1358 continue; 1498 continue;
1359 } 1499 }
1360 } 1500 }
1361 1501
1362 errno = ENAMETOOLONG; 1502 errno = ENAMETOOLONG;
1363 if (res + 1 + len + 1 >= tmp1) 1503 if (res + 1 + len + 1 >= tmp1)
1364 return; 1504 return -1;
1365 1505
1366 /* copy one component */ 1506 /* copy one component */
1367 *res = '/'; 1507 *res = '/';
1368 memcpy (res + 1, beg, len); 1508 memcpy (res + 1, beg, len);
1369 1509
1370 /* zero-terminate, for readlink */ 1510 /* zero-terminate, for readlink */
1371 res [len + 1] = 0; 1511 res [len + 1] = 0;
1372 1512
1373 /* now check if it's a symlink */ 1513 /* now check if it's a symlink */
1374 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1514 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1375 1515
1376 if (linklen < 0) 1516 if (linklen < 0)
1377 { 1517 {
1378 if (errno != EINVAL) 1518 if (errno != EINVAL)
1379 return; 1519 return -1;
1380 1520
1381 /* it's a normal directory. hopefully */ 1521 /* it's a normal directory. hopefully */
1382 res += len + 1; 1522 res += len + 1;
1383 } 1523 }
1384 else 1524 else
1386 /* yay, it was a symlink - build new path in tmp2 */ 1526 /* yay, it was a symlink - build new path in tmp2 */
1387 int rellen = strlen (rel); 1527 int rellen = strlen (rel);
1388 1528
1389 errno = ENAMETOOLONG; 1529 errno = ENAMETOOLONG;
1390 if (linklen + 1 + rellen >= PATH_MAX) 1530 if (linklen + 1 + rellen >= PATH_MAX)
1391 return; 1531 return -1;
1392 1532
1393 errno = ELOOP; 1533 errno = ELOOP;
1394 if (!--symlinks) 1534 if (!--symlinks)
1395 return; 1535 return -1;
1396 1536
1397 if (*tmp1 == '/') 1537 if (*tmp1 == '/')
1398 res = req->ptr2; /* symlink resolves to an absolute path */ 1538 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1399 1539
1400 /* we need to be careful, as rel might point into tmp2 already */ 1540 /* we need to be careful, as rel might point into tmp2 already */
1401 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1541 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1402 tmp2 [linklen] = '/'; 1542 tmp2 [linklen] = '/';
1403 memcpy (tmp2, tmp1, linklen); 1543 memcpy (tmp2, tmp1, linklen);
1405 rel = tmp2; 1545 rel = tmp2;
1406 } 1546 }
1407 } 1547 }
1408 1548
1409 /* special case for the lone root path */ 1549 /* special case for the lone root path */
1410 if (res == req->ptr2) 1550 if (res == tmpbuf->ptr)
1411 *res++ = '/'; 1551 *res++ = '/';
1412 1552
1413 req->result = res - (char *)req->ptr2; 1553 return res - (char *)tmpbuf->ptr;
1414
1415done:
1416 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1417} 1554}
1418 1555
1419static signed char 1556static signed char
1420eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1557eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1421{ 1558{
1429 1566
1430#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1567#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1431#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1568#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1432 1569
1433static void 1570static void
1434eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1571eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1435{ 1572{
1436 unsigned char bits [9 + sizeof (ino_t) * 8]; 1573 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1437 unsigned char *bit = bits; 1574 unsigned char *bit = bits;
1438 1575
1439 assert (CHAR_BIT == 8); 1576 assert (CHAR_BIT == 8);
1440 assert (sizeof (eio_dirent) * 8 < 256); 1577 assert (sizeof (eio_dirent) * 8 < 256);
1441 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1578 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1443 1580
1444 if (size <= EIO_SORT_FAST) 1581 if (size <= EIO_SORT_FAST)
1445 return; 1582 return;
1446 1583
1447 /* first prepare an array of bits to test in our radix sort */ 1584 /* first prepare an array of bits to test in our radix sort */
1448 /* try to take endianness into account, as well as differences in ino_t sizes */ 1585 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1449 /* inode_bits must contain all inodes ORed together */ 1586 /* inode_bits must contain all inodes ORed together */
1450 /* which is used to skip bits that are 0 everywhere, which is very common */ 1587 /* which is used to skip bits that are 0 everywhere, which is very common */
1451 { 1588 {
1452 ino_t endianness; 1589 eio_ino_t endianness;
1453 int i, j; 1590 int i, j;
1454 1591
1455 /* we store the byte offset of byte n into byte n of "endianness" */ 1592 /* we store the byte offset of byte n into byte n of "endianness" */
1456 for (i = 0; i < sizeof (ino_t); ++i) 1593 for (i = 0; i < sizeof (eio_ino_t); ++i)
1457 ((unsigned char *)&endianness)[i] = i; 1594 ((unsigned char *)&endianness)[i] = i;
1458 1595
1459 *bit++ = 0; 1596 *bit++ = 0;
1460 1597
1461 for (i = 0; i < sizeof (ino_t); ++i) 1598 for (i = 0; i < sizeof (eio_ino_t); ++i)
1462 { 1599 {
1463 /* shifting off the byte offsets out of "endianness" */ 1600 /* shifting off the byte offsets out of "endianness" */
1464 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1601 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1465 endianness >>= 8; 1602 endianness >>= 8;
1466 1603
1467 for (j = 0; j < 8; ++j) 1604 for (j = 0; j < 8; ++j)
1468 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1605 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1469 *bit++ = offs + j; 1606 *bit++ = offs + j;
1470 } 1607 }
1471 1608
1472 for (j = 0; j < 8; ++j) 1609 for (j = 0; j < 8; ++j)
1473 if (score_bits & (1 << j)) 1610 if (score_bits & (1 << j))
1474 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1611 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1475 } 1612 }
1476 1613
1477 /* now actually do the sorting (a variant of MSD radix sort) */ 1614 /* now actually do the sorting (a variant of MSD radix sort) */
1478 { 1615 {
1479 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1616 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1480 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1617 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1481 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1618 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1482 int stk_idx = 0; 1619 int stk_idx = 0;
1483 1620
1484 base_stk [stk_idx] = dents; 1621 base_stk [stk_idx] = dents;
1485 end_stk [stk_idx] = dents + size; 1622 end_stk [stk_idx] = dents + size;
1486 bit_stk [stk_idx] = bit - 1; 1623 bit_stk [stk_idx] = bit - 1;
1565 } 1702 }
1566 } 1703 }
1567} 1704}
1568 1705
1569static void 1706static void
1570eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1707eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1571{ 1708{
1572 if (size <= 1) 1709 if (size <= 1)
1573 return; /* our insertion sort relies on size > 0 */ 1710 return; /* our insertion sort relies on size > 0 */
1574 1711
1575 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1712 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1583 1720
1584/* read a full directory */ 1721/* read a full directory */
1585static void 1722static void
1586eio__scandir (eio_req *req, etp_worker *self) 1723eio__scandir (eio_req *req, etp_worker *self)
1587{ 1724{
1588 DIR *dirp;
1589 EIO_STRUCT_DIRENT *entp;
1590 char *name, *names; 1725 char *name, *names;
1591 int namesalloc = 4096; 1726 int namesalloc = 4096 - sizeof (void *) * 4;
1592 int namesoffs = 0; 1727 int namesoffs = 0;
1593 int flags = req->int1; 1728 int flags = req->int1;
1594 eio_dirent *dents = 0; 1729 eio_dirent *dents = 0;
1595 int dentalloc = 128; 1730 int dentalloc = 128;
1596 int dentoffs = 0; 1731 int dentoffs = 0;
1597 ino_t inode_bits = 0; 1732 eio_ino_t inode_bits = 0;
1733#ifdef _WIN32
1734 HANDLE dirp;
1735 WIN32_FIND_DATA entp;
1736#else
1737 DIR *dirp;
1738 EIO_STRUCT_DIRENT *entp;
1739#endif
1598 1740
1599 req->result = -1; 1741 req->result = -1;
1600 1742
1601 if (!(flags & EIO_READDIR_DENTS)) 1743 if (!(flags & EIO_READDIR_DENTS))
1602 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1744 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1603 1745
1604 X_LOCK (wrklock); 1746#ifdef _WIN32
1605 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1747 {
1748 int len = strlen ((const char *)req->ptr1);
1749 char *path = malloc (MAX_PATH);
1750 const char *fmt;
1751 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1752
1753 if (!len)
1754 fmt = "./*";
1755 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1756 fmt = "%s*";
1757 else
1758 fmt = "%s/*";
1759
1760 _snprintf (path, MAX_PATH, fmt, reqpath);
1761 dirp = FindFirstFile (path, &entp);
1762 free (path);
1763
1764 if (dirp == INVALID_HANDLE_VALUE)
1765 {
1766 /* should steal _dosmaperr */
1767 switch (GetLastError ())
1768 {
1769 case ERROR_FILE_NOT_FOUND:
1770 req->result = 0;
1771 break;
1772
1773 case ERROR_INVALID_NAME:
1774 case ERROR_PATH_NOT_FOUND:
1775 case ERROR_NO_MORE_FILES:
1776 errno = ENOENT;
1777 break;
1778
1779 case ERROR_NOT_ENOUGH_MEMORY:
1780 errno = ENOMEM;
1781 break;
1782
1783 default:
1784 errno = EINVAL;
1785 break;
1786 }
1787
1788 return;
1789 }
1790 }
1791#else
1792 #if HAVE_AT
1793 if (req->wd)
1794 {
1795 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1796
1797 if (fd < 0)
1798 return;
1799
1800 dirp = fdopendir (fd);
1801
1802 if (!dirp)
1803 close (fd);
1804 }
1805 else
1606 self->dirp = dirp = opendir (req->ptr1); 1806 dirp = opendir (req->ptr1);
1807 #else
1808 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1809 #endif
1810
1811 if (!dirp)
1812 return;
1813#endif
1607 1814
1608 if (req->flags & EIO_FLAG_PTR1_FREE) 1815 if (req->flags & EIO_FLAG_PTR1_FREE)
1609 free (req->ptr1); 1816 free (req->ptr1);
1610 1817
1611 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1818 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1612 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1819 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1613 req->ptr2 = names = malloc (namesalloc); 1820 req->ptr2 = names = malloc (namesalloc);
1614 X_UNLOCK (wrklock);
1615 1821
1616 if (dirp && names && (!flags || dents)) 1822 if (!names || (flags && !dents))
1823 return;
1824
1617 for (;;) 1825 for (;;)
1618 { 1826 {
1827 int done;
1828
1829#ifdef _WIN32
1830 done = !dirp;
1831#else
1619 errno = 0; 1832 errno = 0;
1620 entp = readdir (dirp); 1833 entp = readdir (dirp);
1834 done = !entp;
1835#endif
1621 1836
1622 if (!entp) 1837 if (done)
1623 { 1838 {
1839#ifndef _WIN32
1840 int old_errno = errno;
1841 closedir (dirp);
1842 errno = old_errno;
1843
1624 if (errno) 1844 if (errno)
1625 break; 1845 break;
1846#endif
1626 1847
1627 /* sort etc. */ 1848 /* sort etc. */
1628 req->int1 = flags; 1849 req->int1 = flags;
1629 req->result = dentoffs; 1850 req->result = dentoffs;
1630 1851
1631 if (flags & EIO_READDIR_STAT_ORDER) 1852 if (flags & EIO_READDIR_STAT_ORDER)
1632 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1853 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1633 else if (flags & EIO_READDIR_DIRS_FIRST) 1854 else if (flags & EIO_READDIR_DIRS_FIRST)
1634 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1855 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1635 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1856 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1636 else 1857 else
1858 {
1859 /* in this case, all is known, and we just put dirs first and sort them */
1860 eio_dirent *oth = dents + dentoffs;
1861 eio_dirent *dir = dents;
1862
1863 /* now partition dirs to the front, and non-dirs to the back */
1864 /* by walking from both sides and swapping if necessary */
1865 while (oth > dir)
1866 {
1867 if (dir->type == EIO_DT_DIR)
1868 ++dir;
1869 else if ((--oth)->type == EIO_DT_DIR)
1870 {
1871 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1872
1873 ++dir;
1874 }
1875 }
1876
1877 /* now sort the dirs only (dirs all have the same score) */
1878 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1879 }
1880
1881 break;
1882 }
1883
1884 /* now add the entry to our list(s) */
1885 name = D_NAME (entp);
1886
1887 /* skip . and .. entries */
1888 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1889 {
1890 int len = D_NAMLEN (entp) + 1;
1891
1892 while (ecb_expect_false (namesoffs + len > namesalloc))
1893 {
1894 namesalloc *= 2;
1895 req->ptr2 = names = realloc (names, namesalloc);
1896
1897 if (!names)
1898 break;
1899 }
1900
1901 memcpy (names + namesoffs, name, len);
1902
1903 if (dents)
1904 {
1905 struct eio_dirent *ent;
1906
1907 if (ecb_expect_false (dentoffs == dentalloc))
1637 { 1908 {
1638 /* in this case, all is known, and we just put dirs first and sort them */ 1909 dentalloc *= 2;
1910 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1911
1912 if (!dents)
1913 break;
1914 }
1915
1639 eio_dirent *oth = dents + dentoffs; 1916 ent = dents + dentoffs;
1640 eio_dirent *dir = dents;
1641 1917
1642 /* now partition dirs to the front, and non-dirs to the back */ 1918 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1643 /* by walking from both sides and swapping if necessary */ 1919 ent->namelen = len - 1;
1644 while (oth > dir) 1920 ent->inode = D_INO (entp);
1921
1922 inode_bits |= ent->inode;
1923
1924 switch (D_TYPE (entp))
1925 {
1926 default:
1927 ent->type = EIO_DT_UNKNOWN;
1928 flags |= EIO_READDIR_FOUND_UNKNOWN;
1929 break;
1930
1931 #ifdef DT_FIFO
1932 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1933 #endif
1934 #ifdef DT_CHR
1935 case DT_CHR: ent->type = EIO_DT_CHR; break;
1936 #endif
1937 #ifdef DT_MPC
1938 case DT_MPC: ent->type = EIO_DT_MPC; break;
1939 #endif
1940 #ifdef DT_DIR
1941 case DT_DIR: ent->type = EIO_DT_DIR; break;
1942 #endif
1943 #ifdef DT_NAM
1944 case DT_NAM: ent->type = EIO_DT_NAM; break;
1945 #endif
1946 #ifdef DT_BLK
1947 case DT_BLK: ent->type = EIO_DT_BLK; break;
1948 #endif
1949 #ifdef DT_MPB
1950 case DT_MPB: ent->type = EIO_DT_MPB; break;
1951 #endif
1952 #ifdef DT_REG
1953 case DT_REG: ent->type = EIO_DT_REG; break;
1954 #endif
1955 #ifdef DT_NWK
1956 case DT_NWK: ent->type = EIO_DT_NWK; break;
1957 #endif
1958 #ifdef DT_CMP
1959 case DT_CMP: ent->type = EIO_DT_CMP; break;
1960 #endif
1961 #ifdef DT_LNK
1962 case DT_LNK: ent->type = EIO_DT_LNK; break;
1963 #endif
1964 #ifdef DT_SOCK
1965 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1966 #endif
1967 #ifdef DT_DOOR
1968 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1969 #endif
1970 #ifdef DT_WHT
1971 case DT_WHT: ent->type = EIO_DT_WHT; break;
1972 #endif
1973 }
1974
1975 ent->score = 7;
1976
1977 if (flags & EIO_READDIR_DIRS_FIRST)
1978 {
1979 if (ent->type == EIO_DT_UNKNOWN)
1645 { 1980 {
1646 if (dir->type == EIO_DT_DIR) 1981 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1647 ++dir; 1982 ent->score = 1;
1648 else if ((--oth)->type == EIO_DT_DIR) 1983 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1649 { 1984 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1650 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1651
1652 ++dir;
1653 }
1654 } 1985 }
1655 1986 else if (ent->type == EIO_DT_DIR)
1656 /* now sort the dirs only (dirs all have the same score) */ 1987 ent->score = 0;
1657 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1658 } 1988 }
1659
1660 break;
1661 }
1662
1663 /* now add the entry to our list(s) */
1664 name = entp->d_name;
1665
1666 /* skip . and .. entries */
1667 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1668 {
1669 int len = D_NAMLEN (entp) + 1;
1670
1671 while (ecb_expect_false (namesoffs + len > namesalloc))
1672 {
1673 namesalloc *= 2;
1674 X_LOCK (wrklock);
1675 req->ptr2 = names = realloc (names, namesalloc);
1676 X_UNLOCK (wrklock);
1677
1678 if (!names)
1679 break;
1680 } 1989 }
1681 1990
1682 memcpy (names + namesoffs, name, len);
1683
1684 if (dents)
1685 {
1686 struct eio_dirent *ent;
1687
1688 if (ecb_expect_false (dentoffs == dentalloc))
1689 {
1690 dentalloc *= 2;
1691 X_LOCK (wrklock);
1692 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1693 X_UNLOCK (wrklock);
1694
1695 if (!dents)
1696 break;
1697 }
1698
1699 ent = dents + dentoffs;
1700
1701 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1702 ent->namelen = len - 1;
1703 ent->inode = D_INO (entp);
1704
1705 inode_bits |= ent->inode;
1706
1707 switch (D_TYPE (entp))
1708 {
1709 default:
1710 ent->type = EIO_DT_UNKNOWN;
1711 flags |= EIO_READDIR_FOUND_UNKNOWN;
1712 break;
1713
1714 #ifdef DT_FIFO
1715 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1716 #endif
1717 #ifdef DT_CHR
1718 case DT_CHR: ent->type = EIO_DT_CHR; break;
1719 #endif
1720 #ifdef DT_MPC
1721 case DT_MPC: ent->type = EIO_DT_MPC; break;
1722 #endif
1723 #ifdef DT_DIR
1724 case DT_DIR: ent->type = EIO_DT_DIR; break;
1725 #endif
1726 #ifdef DT_NAM
1727 case DT_NAM: ent->type = EIO_DT_NAM; break;
1728 #endif
1729 #ifdef DT_BLK
1730 case DT_BLK: ent->type = EIO_DT_BLK; break;
1731 #endif
1732 #ifdef DT_MPB
1733 case DT_MPB: ent->type = EIO_DT_MPB; break;
1734 #endif
1735 #ifdef DT_REG
1736 case DT_REG: ent->type = EIO_DT_REG; break;
1737 #endif
1738 #ifdef DT_NWK
1739 case DT_NWK: ent->type = EIO_DT_NWK; break;
1740 #endif
1741 #ifdef DT_CMP
1742 case DT_CMP: ent->type = EIO_DT_CMP; break;
1743 #endif
1744 #ifdef DT_LNK
1745 case DT_LNK: ent->type = EIO_DT_LNK; break;
1746 #endif
1747 #ifdef DT_SOCK
1748 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1749 #endif
1750 #ifdef DT_DOOR
1751 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1752 #endif
1753 #ifdef DT_WHT
1754 case DT_WHT: ent->type = EIO_DT_WHT; break;
1755 #endif
1756 }
1757
1758 ent->score = 7;
1759
1760 if (flags & EIO_READDIR_DIRS_FIRST)
1761 {
1762 if (ent->type == EIO_DT_UNKNOWN)
1763 {
1764 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1765 ent->score = 1;
1766 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1767 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1768 }
1769 else if (ent->type == EIO_DT_DIR)
1770 ent->score = 0;
1771 }
1772 }
1773
1774 namesoffs += len; 1991 namesoffs += len;
1775 ++dentoffs; 1992 ++dentoffs;
1776 } 1993 }
1777 1994
1778 if (EIO_CANCELLED (req)) 1995 if (EIO_CANCELLED (req))
1779 { 1996 {
1780 errno = ECANCELED; 1997 errno = ECANCELED;
1781 break; 1998 break;
1782 } 1999 }
2000
2001#ifdef _WIN32
2002 if (!FindNextFile (dirp, &entp))
2003 {
2004 FindClose (dirp);
2005 dirp = 0;
1783 } 2006 }
2007#endif
2008 }
1784} 2009}
2010
2011/*****************************************************************************/
2012/* working directory stuff */
2013/* various deficiencies in the posix 2008 api force us to */
2014/* keep the absolute path in string form at all times */
2015/* fuck yeah. */
2016
2017#if !HAVE_AT
2018
2019/* a bit like realpath, but usually faster because it doesn'T have to return */
2020/* an absolute or canonical path */
2021static const char *
2022wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2023{
2024 if (!wd || *path == '/')
2025 return path;
2026
2027 if (path [0] == '.' && !path [1])
2028 return wd->str;
2029
2030 {
2031 int l1 = wd->len;
2032 int l2 = strlen (path);
2033
2034 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2035
2036 memcpy (res, wd->str, l1);
2037 res [l1] = '/';
2038 memcpy (res + l1 + 1, path, l2 + 1);
2039
2040 return res;
2041 }
2042}
2043
2044#endif
2045
2046static eio_wd
2047eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2048{
2049 int fd;
2050 eio_wd res;
2051 int len = eio__realpath (tmpbuf, wd, path);
2052
2053 if (len < 0)
2054 return EIO_INVALID_WD;
2055
2056#if HAVE_AT
2057 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2058
2059 if (fd < 0)
2060 return EIO_INVALID_WD;
2061#endif
2062
2063 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2064
2065#if HAVE_AT
2066 res->fd = fd;
2067#endif
2068
2069 res->len = len;
2070 memcpy (res->str, tmpbuf->ptr, len);
2071 res->str [len] = 0;
2072
2073 return res;
2074}
2075
2076eio_wd
2077eio_wd_open_sync (eio_wd wd, const char *path)
2078{
2079 struct tmpbuf tmpbuf = { 0 };
2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
2082
2083 return wd;
2084}
2085
2086void
2087eio_wd_close_sync (eio_wd wd)
2088{
2089 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2090 {
2091 #if HAVE_AT
2092 close (wd->fd);
2093 #endif
2094 free (wd);
2095 }
2096}
2097
2098#if HAVE_AT
2099
2100/* they forgot these */
2101
2102static int
2103eio__truncateat (int dirfd, const char *path, off_t length)
2104{
2105 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = ftruncate (fd, length);
2112 close (fd);
2113 return res;
2114}
2115
2116static int
2117eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2118{
2119 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2120 int res;
2121
2122 if (fd < 0)
2123 return fd;
2124
2125 res = fstatvfs (fd, buf);
2126 close (fd);
2127 return res;
2128
2129}
2130
2131#endif
1785 2132
1786/*****************************************************************************/ 2133/*****************************************************************************/
1787 2134
1788#define ALLOC(len) \ 2135#define ALLOC(len) \
1789 if (!req->ptr2) \ 2136 if (!req->ptr2) \
1798 req->result = -1; \ 2145 req->result = -1; \
1799 break; \ 2146 break; \
1800 } \ 2147 } \
1801 } 2148 }
1802 2149
2150static void ecb_noinline ecb_cold
2151etp_proc_init (void)
2152{
2153#if HAVE_PRCTL_SET_NAME
2154 /* provide a more sensible "thread name" */
2155 char name[16 + 1];
2156 const int namelen = sizeof (name) - 1;
2157 int len;
2158
2159 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2160 name [namelen] = 0;
2161 len = strlen (name);
2162 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2163 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2164#endif
2165}
2166
1803X_THREAD_PROC (etp_proc) 2167X_THREAD_PROC (etp_proc)
1804{ 2168{
1805 ETP_REQ *req; 2169 ETP_REQ *req;
1806 struct timespec ts; 2170 struct timespec ts;
1807 etp_worker *self = (etp_worker *)thr_arg; 2171 etp_worker *self = (etp_worker *)thr_arg;
1808 2172
2173 etp_proc_init ();
2174
1809 /* try to distribute timeouts somewhat randomly */ 2175 /* try to distribute timeouts somewhat evenly */
1810 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2176 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1811 2177
1812 for (;;) 2178 for (;;)
1813 { 2179 {
2180 ts.tv_sec = 0;
2181
1814 X_LOCK (reqlock); 2182 X_LOCK (reqlock);
1815 2183
1816 for (;;) 2184 for (;;)
1817 { 2185 {
1818 self->req = req = reqq_shift (&req_queue); 2186 req = reqq_shift (&req_queue);
1819 2187
1820 if (req) 2188 if (req)
1821 break; 2189 break;
1822 2190
2191 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2192 {
2193 X_UNLOCK (reqlock);
2194 X_LOCK (wrklock);
2195 --started;
2196 X_UNLOCK (wrklock);
2197 goto quit;
2198 }
2199
1823 ++idle; 2200 ++idle;
1824 2201
1825 ts.tv_sec = time (0) + idle_timeout; 2202 if (idle <= max_idle)
1826 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2203 /* we are allowed to idle, so do so without any timeout */
2204 X_COND_WAIT (reqwait, reqlock);
2205 else
1827 { 2206 {
1828 if (idle > max_idle) 2207 /* initialise timeout once */
1829 { 2208 if (!ts.tv_sec)
1830 --idle; 2209 ts.tv_sec = time (0) + idle_timeout;
1831 X_UNLOCK (reqlock);
1832 X_LOCK (wrklock);
1833 --started;
1834 X_UNLOCK (wrklock);
1835 goto quit;
1836 }
1837 2210
1838 /* we are allowed to idle, so do so without any timeout */
1839 X_COND_WAIT (reqwait, reqlock); 2211 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2212 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1840 } 2213 }
1841 2214
1842 --idle; 2215 --idle;
1843 } 2216 }
1844 2217
1856 ++npending; 2229 ++npending;
1857 2230
1858 if (!reqq_push (&res_queue, req) && want_poll_cb) 2231 if (!reqq_push (&res_queue, req) && want_poll_cb)
1859 want_poll_cb (); 2232 want_poll_cb ();
1860 2233
1861 self->req = 0;
1862 etp_worker_clear (self); 2234 etp_worker_clear (self);
1863 2235
1864 X_UNLOCK (reslock); 2236 X_UNLOCK (reslock);
1865 } 2237 }
1866 2238
1867quit: 2239quit:
2240 free (req);
2241
1868 X_LOCK (wrklock); 2242 X_LOCK (wrklock);
1869 etp_worker_free (self); 2243 etp_worker_free (self);
1870 X_UNLOCK (wrklock); 2244 X_UNLOCK (wrklock);
1871
1872 return 0;
1873} 2245}
1874 2246
1875/*****************************************************************************/ 2247/*****************************************************************************/
1876 2248
1877int ecb_cold 2249int ecb_cold
1878eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2250eio_init (void (*want_poll)(void), void (*done_poll)(void))
1879{ 2251{
2252#if !HAVE_PREADWRITE
2253 X_MUTEX_CREATE (preadwritelock);
2254#endif
2255
1880 return etp_init (want_poll, done_poll); 2256 return etp_init (want_poll, done_poll);
1881} 2257}
1882 2258
1883ecb_inline void 2259ecb_inline void
1884eio_api_destroy (eio_req *req) 2260eio_api_destroy (eio_req *req)
1911 } 2287 }
1912 2288
1913static void 2289static void
1914eio_execute (etp_worker *self, eio_req *req) 2290eio_execute (etp_worker *self, eio_req *req)
1915{ 2291{
2292#if HAVE_AT
2293 int dirfd;
2294#else
2295 const char *path;
2296#endif
2297
1916 if (ecb_expect_false (EIO_CANCELLED (req))) 2298 if (ecb_expect_false (EIO_CANCELLED (req)))
1917 { 2299 {
1918 req->result = -1; 2300 req->result = -1;
1919 req->errorno = ECANCELED; 2301 req->errorno = ECANCELED;
1920 return; 2302 return;
1921 } 2303 }
1922 2304
2305 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2306 {
2307 req->result = -1;
2308 req->errorno = ENOENT;
2309 return;
2310 }
2311
2312 if (req->type >= EIO_OPEN)
2313 {
2314 #if HAVE_AT
2315 dirfd = WD2FD (req->wd);
2316 #else
2317 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2318 #endif
2319 }
2320
1923 switch (req->type) 2321 switch (req->type)
1924 { 2322 {
2323 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2324 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2325 break;
2326 case EIO_WD_CLOSE: req->result = 0;
2327 eio_wd_close_sync (req->wd); break;
2328
2329 case EIO_SEEK: eio__lseek (req); break;
1925 case EIO_READ: ALLOC (req->size); 2330 case EIO_READ: ALLOC (req->size);
1926 req->result = req->offs >= 0 2331 req->result = req->offs >= 0
1927 ? pread (req->int1, req->ptr2, req->size, req->offs) 2332 ? pread (req->int1, req->ptr2, req->size, req->offs)
1928 : read (req->int1, req->ptr2, req->size); break; 2333 : read (req->int1, req->ptr2, req->size); break;
1929 case EIO_WRITE: req->result = req->offs >= 0 2334 case EIO_WRITE: req->result = req->offs >= 0
1930 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2335 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1931 : write (req->int1, req->ptr2, req->size); break; 2336 : write (req->int1, req->ptr2, req->size); break;
1932 2337
1933 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2338 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1934 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2339 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2340
2341#if HAVE_AT
1935 2342
1936 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1937 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2344 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1938 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2347 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2348 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2349 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2350 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2353 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2354 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1939 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
1940 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1941 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1942 2445
1943 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1944 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1945 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1946 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1947 2448
1948 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1949 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1950 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1951 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1952 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1953 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1954 2452
1955 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1956 case EIO_CLOSE: req->result = close (req->int1); break; 2453 case EIO_CLOSE: req->result = close (req->int1); break;
1957 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1958 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1959 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1960 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1961 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1962 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1963 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1964 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1965
1966 case EIO_REALPATH: eio__realpath (req, self); break;
1967
1968 case EIO_READLINK: ALLOC (PATH_MAX);
1969 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1970
1971 case EIO_SYNC: req->result = 0; sync (); break; 2455 case EIO_SYNC: req->result = 0; sync (); break;
1972 case EIO_FSYNC: req->result = fsync (req->int1); break; 2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
1973 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1974 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1975 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1976 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1977 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1978 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1979 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1980 2465
1981 case EIO_READDIR: eio__scandir (req, self); break; 2466 case EIO_READDIR: eio__scandir (req, self); break;
1982 2467
1983 case EIO_BUSY: 2468 case EIO_BUSY:
1993 req->result = select (0, 0, 0, 0, &tv); 2478 req->result = select (0, 0, 0, 0, &tv);
1994 } 2479 }
1995#endif 2480#endif
1996 break; 2481 break;
1997 2482
1998 case EIO_UTIME:
1999 case EIO_FUTIME:
2000 {
2001 struct timeval tv[2];
2002 struct timeval *times;
2003
2004 if (req->nv1 != -1. || req->nv2 != -1.)
2005 {
2006 tv[0].tv_sec = req->nv1;
2007 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2008 tv[1].tv_sec = req->nv2;
2009 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2010
2011 times = tv;
2012 }
2013 else
2014 times = 0;
2015
2016 req->result = req->type == EIO_FUTIME
2017 ? futimes (req->int1, times)
2018 : utimes (req->ptr1, times);
2019 }
2020 break;
2021
2022 case EIO_GROUP: 2483 case EIO_GROUP:
2023 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
2024 2485
2025 case EIO_NOP: 2486 case EIO_NOP:
2026 req->result = 0; 2487 req->result = 0;
2029 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
2030 req->feed (req); 2491 req->feed (req);
2031 break; 2492 break;
2032 2493
2033 default: 2494 default:
2034 errno = ENOSYS;
2035 req->result = -1; 2495 req->result = EIO_ENOSYS ();
2036 break; 2496 break;
2037 } 2497 }
2038 2498
2039 req->errorno = errno; 2499 req->errorno = errno;
2040} 2500}
2041 2501
2042#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
2043 2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2044eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2045{ 2515{
2046 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
2047} 2517}
2048 2518
2064eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2065{ 2535{
2066 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2067} 2537}
2068 2538
2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2542}
2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2069eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2070{ 2555{
2071 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2072} 2557}
2073 2558
2079eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2080{ 2565{
2081 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2082} 2567}
2083 2568
2084eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2087}
2088
2089eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2090{ 2570{
2091 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2092} 2572}
2093 2573
2094eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2095{
2096 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2097}
2098
2099eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2100{ 2575{
2101 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2102} 2577}
2103 2578
2104eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2105{ 2580{
2106 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2107} 2582}
2108 2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
2109eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2110{ 2590{
2111 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2112} 2592}
2113 2593
2139eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2140{ 2620{
2141 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2142} 2622}
2143 2623
2144eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2145{ 2625{
2146 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2147} 2627}
2148 2628
2149eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2169eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2170{ 2650{
2171 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2172} 2652}
2173 2653
2174eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2175{ 2655{
2176 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2177} 2657}
2178 2658
2179eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2326} 2806}
2327 2807
2328/*****************************************************************************/ 2808/*****************************************************************************/
2329/* misc garbage */ 2809/* misc garbage */
2330 2810
2331ssize_t 2811eio_ssize_t
2332eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2333{ 2813{
2334 etp_worker wrk;
2335 ssize_t ret;
2336
2337 wrk.dbuf = 0;
2338
2339 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
2340
2341 if (wrk.dbuf)
2342 free (wrk.dbuf);
2343
2344 return ret;
2345} 2815}
2346 2816

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines