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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 UTC vs.
Revision 1.123 by root, Tue Oct 9 04:53:53 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 #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 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
91#else 232#else
92 233
93# include <sys/time.h> 234 #include <sys/time.h>
94# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
95# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
242 #include <sys/mman.h>
243 #endif
244
245 #define D_NAME(entp) entp->d_name
246
247 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
248 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
249 #define _DIRENT_HAVE_D_TYPE /* sigh */
250 #define D_INO(de) (de)->d_fileno
251 #define D_NAMLEN(de) (de)->d_namlen
252 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
253 #define D_INO(de) (de)->d_ino
254 #endif
255
256 #ifdef _D_EXACT_NAMLEN
257 #undef D_NAMLEN
258 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
259 #endif
260
261 #ifdef _DIRENT_HAVE_D_TYPE
262 #define D_TYPE(de) (de)->d_type
263 #endif
264
265 #ifndef EIO_STRUCT_DIRENT
266 #define EIO_STRUCT_DIRENT struct dirent
267 #endif
268
269#endif
270
271#if HAVE_UTIMES
96# include <utime.h> 272# 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 273#endif
112 274
113#ifdef _D_EXACT_NAMLEN 275#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 276# 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 277#endif
121 278
122# ifndef EIO_STRUCT_DIRENT 279#if HAVE_SYS_PRCTL_H
123# define EIO_STRUCT_DIRENT struct dirent 280# include <sys/prctl.h>
124# endif
125
126#endif 281#endif
127 282
128#if HAVE_SENDFILE 283#if HAVE_SENDFILE
129# if __linux 284# if __linux
130# include <sys/sendfile.h> 285# include <sys/sendfile.h>
145#endif 300#endif
146#ifndef D_INO 301#ifndef D_INO
147# define D_INO(de) 0 302# define D_INO(de) 0
148#endif 303#endif
149#ifndef D_NAMLEN 304#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 305# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 306#endif
152 307
153/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 309#ifndef NAME_MAX
155# define NAME_MAX 4096 310# define NAME_MAX 4096
162 317
163/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
165 320
166#define dBUF \ 321#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 323 errno = ENOMEM; \
172 if (!eio_buf) \ 324 if (!eio_buf) \
173 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
174 329
175#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
331
332/*****************************************************************************/
333
334struct tmpbuf
335{
336 void *ptr;
337 int len;
338};
339
340static void *
341tmpbuf_get (struct tmpbuf *buf, int len)
342{
343 if (buf->len < len)
344 {
345 free (buf->ptr);
346 buf->ptr = malloc (buf->len = len);
347 }
348
349 return buf->ptr;
350}
351
352struct 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 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/*****************************************************************************/
176 375
177#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
179 378
180struct etp_worker; 379struct etp_worker;
184static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 387
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/*****************************************************************************/ 388/*****************************************************************************/
207 389
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 391
210/* calculate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
221static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
222 404
223static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
225 407
226static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 411static 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 */ 412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231 413
232static xmutex_t wrklock; 414static xmutex_t wrklock;
233static xmutex_t reslock; 415static xmutex_t reslock;
234static xmutex_t reqlock; 416static xmutex_t reqlock;
235static xcond_t reqwait; 417static xcond_t reqwait;
236 418
237#if !HAVE_PREADWRITE
238/*
239 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine.
242 */
243static xmutex_t preadwritelock = X_MUTEX_INIT;
244#endif
245
246typedef struct etp_worker 419typedef struct etp_worker
247{ 420{
421 struct tmpbuf tmpbuf;
422
248 /* locked by wrklock */ 423 /* locked by wrklock */
249 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
250 425
251 xthread_t tid; 426 xthread_t tid;
252 427
253 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
257} etp_worker; 431} etp_worker;
258 432
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
260 434
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263 437
264/* worker threads management */ 438/* worker threads management */
265 439
266static void ecb_cold 440static void
267etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
268{ 442{
269 ETP_WORKER_CLEAR (wrk);
270} 443}
271 444
272static void ecb_cold 445static void ecb_cold
273etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
274{ 447{
448 free (wrk->tmpbuf.ptr);
449
275 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
277 452
278 free (wrk); 453 free (wrk);
279} 454}
335} etp_reqq; 510} etp_reqq;
336 511
337static etp_reqq req_queue; 512static etp_reqq req_queue;
338static etp_reqq res_queue; 513static etp_reqq res_queue;
339 514
515static void ecb_noinline ecb_cold
516reqq_init (etp_reqq *q)
517{
518 int pri;
519
520 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
521 q->qs[pri] = q->qe[pri] = 0;
522
523 q->size = 0;
524}
525
340static int ecb_noinline 526static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
342{ 528{
343 int pri = req->pri; 529 int pri = req->pri;
344 req->next = 0; 530 req->next = 0;
378 } 564 }
379 565
380 abort (); 566 abort ();
381} 567}
382 568
383static void ecb_cold 569static int ecb_cold
384etp_thread_init (void) 570etp_init (void (*want_poll)(void), void (*done_poll)(void))
385{ 571{
386#if !HAVE_PREADWRITE
387 X_MUTEX_CREATE (preadwritelock);
388#endif
389 X_MUTEX_CREATE (wrklock); 572 X_MUTEX_CREATE (wrklock);
390 X_MUTEX_CREATE (reslock); 573 X_MUTEX_CREATE (reslock);
391 X_MUTEX_CREATE (reqlock); 574 X_MUTEX_CREATE (reqlock);
392 X_COND_CREATE (reqwait); 575 X_COND_CREATE (reqwait);
393}
394 576
395static void ecb_cold 577 reqq_init (&req_queue);
396etp_atfork_prepare (void) 578 reqq_init (&res_queue);
397{
398}
399 579
400static void ecb_cold 580 wrk_first.next =
401etp_atfork_parent (void) 581 wrk_first.prev = &wrk_first;
402{
403}
404
405static void ecb_cold
406etp_atfork_child (void)
407{
408 ETP_REQ *prv;
409
410 while ((prv = reqq_shift (&req_queue)))
411 ETP_DESTROY (prv);
412
413 while ((prv = reqq_shift (&res_queue)))
414 ETP_DESTROY (prv);
415
416 while (wrk_first.next != &wrk_first)
417 {
418 etp_worker *wrk = wrk_first.next;
419
420 if (wrk->req)
421 ETP_DESTROY (wrk->req);
422
423 etp_worker_clear (wrk);
424 etp_worker_free (wrk);
425 }
426 582
427 started = 0; 583 started = 0;
428 idle = 0; 584 idle = 0;
429 nreqs = 0; 585 nreqs = 0;
430 nready = 0; 586 nready = 0;
431 npending = 0; 587 npending = 0;
432
433 etp_thread_init ();
434}
435
436static void ecb_cold
437etp_once_init (void)
438{
439 etp_thread_init ();
440 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
441}
442
443static int ecb_cold
444etp_init (void (*want_poll)(void), void (*done_poll)(void))
445{
446 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
447
448 pthread_once (&doinit, etp_once_init);
449 588
450 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
451 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
452 591
453 return 0; 592 return 0;
493} 632}
494 633
495static void ecb_cold 634static void ecb_cold
496etp_end_thread (void) 635etp_end_thread (void)
497{ 636{
498 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
499 638
500 req->type = -1; 639 req->type = -1;
501 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
502 641
503 X_LOCK (reqlock); 642 X_LOCK (reqlock);
830} 969}
831 970
832/*****************************************************************************/ 971/*****************************************************************************/
833/* work around various missing functions */ 972/* work around various missing functions */
834 973
835#if !HAVE_PREADWRITE
836# undef pread
837# undef pwrite
838# define pread eio__pread
839# define pwrite eio__pwrite
840
841static ssize_t
842eio__pread (int fd, void *buf, size_t count, off_t offset)
843{
844 ssize_t res;
845 off_t ooffset;
846
847 X_LOCK (preadwritelock);
848 ooffset = lseek (fd, 0, SEEK_CUR);
849 lseek (fd, offset, SEEK_SET);
850 res = read (fd, buf, count);
851 lseek (fd, ooffset, SEEK_SET);
852 X_UNLOCK (preadwritelock);
853
854 return res;
855}
856
857static ssize_t
858eio__pwrite (int fd, void *buf, size_t count, off_t offset)
859{
860 ssize_t res;
861 off_t ooffset;
862
863 X_LOCK (preadwritelock);
864 ooffset = lseek (fd, 0, SEEK_CUR);
865 lseek (fd, offset, SEEK_SET);
866 res = write (fd, buf, count);
867 lseek (fd, ooffset, SEEK_SET);
868 X_UNLOCK (preadwritelock);
869
870 return res;
871}
872#endif
873
874#ifndef HAVE_UTIMES 974#ifndef HAVE_UTIMES
875 975
876# undef utimes 976# undef utimes
877# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
878 978
910 1010
911#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
912# undef fdatasync 1012# undef fdatasync
913# define fdatasync(fd) fsync (fd) 1013# define fdatasync(fd) fsync (fd)
914#endif 1014#endif
1015
1016static int
1017eio__syncfs (int fd)
1018{
1019 int res;
1020
1021#if HAVE_SYS_SYNCFS
1022 res = (int)syscall (__NR_syncfs, (int)(fd));
1023#else
1024 res = EIO_ENOSYS ();
1025#endif
1026
1027 if (res < 0 && errno == ENOSYS && fd >= 0)
1028 sync ();
1029
1030 return res;
1031}
915 1032
916/* sync_file_range always needs emulation */ 1033/* sync_file_range always needs emulation */
917static int 1034static int
918eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1035eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
919{ 1036{
942} 1059}
943 1060
944static int 1061static int
945eio__fallocate (int fd, int mode, off_t offset, size_t len) 1062eio__fallocate (int fd, int mode, off_t offset, size_t len)
946{ 1063{
947#if HAVE_FALLOCATE 1064#if HAVE_LINUX_FALLOCATE
948 return fallocate (fd, mode, offset, len); 1065 return fallocate (fd, mode, offset, len);
949#else 1066#else
950 errno = ENOSYS; 1067 return EIO_ENOSYS ();
951 return -1;
952#endif 1068#endif
953} 1069}
954 1070
955#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
956# undef readahead 1072# undef readahead
957# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
958 1074
959static ssize_t 1075static eio_ssize_t
960eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1076eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
961{ 1077{
962 size_t todo = count; 1078 size_t todo = count;
963 dBUF; 1079 dBUF;
964 1080
969 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
970 offset += len; 1086 offset += len;
971 todo -= len; 1087 todo -= len;
972 } 1088 }
973 1089
974 errno = 0; 1090 FUBd;
1091
1092 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1093 /* but not for e.g. EIO or eof, so we also never fail */
975 return count; 1094 return 0;
976} 1095}
977 1096
978#endif 1097#endif
979 1098
980/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
981static ssize_t 1100static eio_ssize_t
982eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1101eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
983{ 1102{
984 ssize_t written = 0; 1103 eio_ssize_t written = 0;
985 ssize_t res; 1104 eio_ssize_t res;
986 1105
987 if (!count) 1106 if (!count)
988 return 0; 1107 return 0;
989 1108
990 for (;;) 1109 for (;;)
1015 1134
1016 /* according to source inspection, this is correct, and useful behaviour */ 1135 /* according to source inspection, this is correct, and useful behaviour */
1017 if (sbytes) 1136 if (sbytes)
1018 res = sbytes; 1137 res = sbytes;
1019 1138
1020# elif defined (__APPLE__) 1139# elif defined __APPLE__
1021 off_t sbytes = count; 1140 off_t sbytes = count;
1022 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1023 1142
1024 /* according to the manpage, sbytes is always valid */ 1143 /* according to the manpage, sbytes is always valid */
1025 if (sbytes) 1144 if (sbytes)
1042 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
1043 res = sbytes; 1162 res = sbytes;
1044 1163
1045# endif 1164# endif
1046 1165
1047#elif defined (_WIN32) 1166#elif defined (_WIN32) && 0
1048 /* does not work, just for documentation of what would need to be done */ 1167 /* does not work, just for documentation of what would need to be done */
1049 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1168 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1050 /* libeio guarantees that the file offset does not change, and windows */ 1169 /* libeio guarantees that the file offset does not change, and windows */
1051 /* has no way to get an independent handle to the same file description */ 1170 /* has no way to get an independent handle to the same file description */
1052 HANDLE h = TO_SOCKET (ifd); 1171 HANDLE h = TO_SOCKET (ifd);
1053 SetFilePointer (h, offset, 0, FILE_BEGIN); 1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1054 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1055 1174
1056#else 1175#else
1057 res = -1; 1176 res = EIO_ENOSYS ();
1058 errno = ENOSYS;
1059#endif 1177#endif
1060 1178
1061 /* we assume sendfile can copy at least 128mb in one go */ 1179 /* we assume sendfile can copy at least 128mb in one go */
1062 if (res <= 128 * 1024 * 1024) 1180 if (res <= 128 * 1024 * 1024)
1063 { 1181 {
1085 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1086 /* BSDs */ 1204 /* BSDs */
1087#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1205#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1088 || errno == ENOTSUP 1206 || errno == ENOTSUP
1089#endif 1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1090 || errno == EOPNOTSUPP /* BSDs */ 1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
1091#if __solaris 1211#if __solaris
1092 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1093#endif 1213#endif
1094 ) 1214 )
1095 ) 1215 )
1099 1219
1100 res = 0; 1220 res = 0;
1101 1221
1102 while (count) 1222 while (count)
1103 { 1223 {
1104 ssize_t cnt; 1224 eio_ssize_t cnt;
1105 1225
1106 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1226 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1107 1227
1108 if (cnt <= 0) 1228 if (cnt <= 0)
1109 { 1229 {
1121 1241
1122 offset += cnt; 1242 offset += cnt;
1123 res += cnt; 1243 res += cnt;
1124 count -= cnt; 1244 count -= cnt;
1125 } 1245 }
1246
1247 FUBd;
1126 } 1248 }
1127 1249
1128 return res; 1250 return res;
1129} 1251}
1130 1252
1157 /* round up length */ 1279 /* round up length */
1158 *length = (*length + mask) & ~mask; 1280 *length = (*length + mask) & ~mask;
1159} 1281}
1160 1282
1161#if !_POSIX_MEMLOCK 1283#if !_POSIX_MEMLOCK
1162# define eio__mlockall(a) ((errno = ENOSYS), -1) 1284# define eio__mlockall(a) EIO_ENOSYS ()
1163#else 1285#else
1164 1286
1165static int 1287static int
1166eio__mlockall (int flags) 1288eio__mlockall (int flags)
1167{ 1289{
1181 return mlockall (flags); 1303 return mlockall (flags);
1182} 1304}
1183#endif 1305#endif
1184 1306
1185#if !_POSIX_MEMLOCK_RANGE 1307#if !_POSIX_MEMLOCK_RANGE
1186# define eio__mlock(a,b) ((errno = ENOSYS), -1) 1308# define eio__mlock(a,b) EIO_ENOSYS ()
1187#else 1309#else
1188 1310
1189static int 1311static int
1190eio__mlock (void *addr, size_t length) 1312eio__mlock (void *addr, size_t length)
1191{ 1313{
1195} 1317}
1196 1318
1197#endif 1319#endif
1198 1320
1199#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1321#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1200# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1322# define eio__msync(a,b,c) EIO_ENOSYS ()
1201#else 1323#else
1202 1324
1203static int 1325static int
1204eio__msync (void *mem, size_t len, int flags) 1326eio__msync (void *mem, size_t len, int flags)
1205{ 1327{
1246 1368
1247/*****************************************************************************/ 1369/*****************************************************************************/
1248/* requests implemented outside eio_execute, because they are so large */ 1370/* requests implemented outside eio_execute, because they are so large */
1249 1371
1250static void 1372static void
1251eio__realpath (eio_req *req, etp_worker *self) 1373eio__lseek (eio_req *req)
1252{ 1374{
1253 char *rel = req->ptr1; 1375 /* this usually gets optimised away completely, or your compiler sucks, */
1376 /* or the whence constants really are not 0, 1, 2 */
1377 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1378 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1379 : req->int2 == EIO_SEEK_END ? SEEK_END
1380 : req->int2;
1381
1382 req->offs = lseek (req->int1, req->offs, whence);
1383 req->result = req->offs == (off_t)-1 ? -1 : 0;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1254 char *res; 1391 char *res;
1255 char *tmp1, *tmp2; 1392 char *tmp1, *tmp2;
1256#if SYMLOOP_MAX > 32 1393#if SYMLOOP_MAX > 32
1257 int symlinks = SYMLOOP_MAX; 1394 int symlinks = SYMLOOP_MAX;
1258#else 1395#else
1259 int symlinks = 32; 1396 int symlinks = 32;
1260#endif 1397#endif
1261 1398
1262 req->result = -1;
1263
1264 errno = EINVAL; 1399 errno = EINVAL;
1265 if (!rel) 1400 if (!rel)
1266 return; 1401 return -1;
1267 1402
1268 errno = ENOENT; 1403 errno = ENOENT;
1269 if (!*rel) 1404 if (!*rel)
1270 return; 1405 return -1;
1271 1406
1272 if (!req->ptr2) 1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1273 {
1274 X_LOCK (wrklock);
1275 req->flags |= EIO_FLAG_PTR2_FREE;
1276 X_UNLOCK (wrklock);
1277 req->ptr2 = malloc (PATH_MAX * 3);
1278
1279 errno = ENOMEM;
1280 if (!req->ptr2)
1281 return;
1282 }
1283
1284 res = req->ptr2;
1285 tmp1 = res + PATH_MAX; 1408 tmp1 = res + PATH_MAX;
1286 tmp2 = tmp1 + PATH_MAX; 1409 tmp2 = tmp1 + PATH_MAX;
1287 1410
1288#if 0 /* disabled, the musl way to do things is just too racy */ 1411#if 0 /* disabled, the musl way to do things is just too racy */
1289#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1308#endif 1431#endif
1309#endif 1432#endif
1310 1433
1311 if (*rel != '/') 1434 if (*rel != '/')
1312 { 1435 {
1436 int len;
1437
1438 errno = ENOENT;
1439 if (wd == EIO_INVALID_WD)
1440 return -1;
1441
1442 if (wd == EIO_CWD)
1443 {
1313 if (!getcwd (res, PATH_MAX)) 1444 if (!getcwd (res, PATH_MAX))
1314 return; 1445 return -1;
1446
1447 len = strlen (res);
1448 }
1449 else
1450 memcpy (res, wd->str, len = wd->len);
1315 1451
1316 if (res [1]) /* only use if not / */ 1452 if (res [1]) /* only use if not / */
1317 res += strlen (res); 1453 res += len;
1318 } 1454 }
1319 1455
1320 while (*rel) 1456 while (*rel)
1321 { 1457 {
1322 ssize_t len, linklen; 1458 eio_ssize_t len, linklen;
1323 char *beg = rel; 1459 const char *beg = rel;
1324 1460
1325 while (*rel && *rel != '/') 1461 while (*rel && *rel != '/')
1326 ++rel; 1462 ++rel;
1327 1463
1328 len = rel - beg; 1464 len = rel - beg;
1340 1476
1341 if (beg [1] == '.' && len == 2) 1477 if (beg [1] == '.' && len == 2)
1342 { 1478 {
1343 /* .. - back up one component, if possible */ 1479 /* .. - back up one component, if possible */
1344 1480
1345 while (res != req->ptr2) 1481 while (res != tmpbuf->ptr)
1346 if (*--res == '/') 1482 if (*--res == '/')
1347 break; 1483 break;
1348 1484
1349 continue; 1485 continue;
1350 } 1486 }
1351 } 1487 }
1352 1488
1353 errno = ENAMETOOLONG; 1489 errno = ENAMETOOLONG;
1354 if (res + 1 + len + 1 >= tmp1) 1490 if (res + 1 + len + 1 >= tmp1)
1355 return; 1491 return -1;
1356 1492
1357 /* copy one component */ 1493 /* copy one component */
1358 *res = '/'; 1494 *res = '/';
1359 memcpy (res + 1, beg, len); 1495 memcpy (res + 1, beg, len);
1360 1496
1361 /* zero-terminate, for readlink */ 1497 /* zero-terminate, for readlink */
1362 res [len + 1] = 0; 1498 res [len + 1] = 0;
1363 1499
1364 /* now check if it's a symlink */ 1500 /* now check if it's a symlink */
1365 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1501 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1366 1502
1367 if (linklen < 0) 1503 if (linklen < 0)
1368 { 1504 {
1369 if (errno != EINVAL) 1505 if (errno != EINVAL)
1370 return; 1506 return -1;
1371 1507
1372 /* it's a normal directory. hopefully */ 1508 /* it's a normal directory. hopefully */
1373 res += len + 1; 1509 res += len + 1;
1374 } 1510 }
1375 else 1511 else
1377 /* yay, it was a symlink - build new path in tmp2 */ 1513 /* yay, it was a symlink - build new path in tmp2 */
1378 int rellen = strlen (rel); 1514 int rellen = strlen (rel);
1379 1515
1380 errno = ENAMETOOLONG; 1516 errno = ENAMETOOLONG;
1381 if (linklen + 1 + rellen >= PATH_MAX) 1517 if (linklen + 1 + rellen >= PATH_MAX)
1382 return; 1518 return -1;
1383 1519
1384 errno = ELOOP; 1520 errno = ELOOP;
1385 if (!--symlinks) 1521 if (!--symlinks)
1386 return; 1522 return -1;
1387 1523
1388 if (*tmp1 == '/') 1524 if (*tmp1 == '/')
1389 res = req->ptr2; /* symlink resolves to an absolute path */ 1525 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1390 1526
1391 /* we need to be careful, as rel might point into tmp2 already */ 1527 /* we need to be careful, as rel might point into tmp2 already */
1392 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1528 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1393 tmp2 [linklen] = '/'; 1529 tmp2 [linklen] = '/';
1394 memcpy (tmp2, tmp1, linklen); 1530 memcpy (tmp2, tmp1, linklen);
1396 rel = tmp2; 1532 rel = tmp2;
1397 } 1533 }
1398 } 1534 }
1399 1535
1400 /* special case for the lone root path */ 1536 /* special case for the lone root path */
1401 if (res == req->ptr2) 1537 if (res == tmpbuf->ptr)
1402 *res++ = '/'; 1538 *res++ = '/';
1403 1539
1404 req->result = res - (char *)req->ptr2; 1540 return res - (char *)tmpbuf->ptr;
1405
1406done:
1407 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1408} 1541}
1409 1542
1410static signed char 1543static signed char
1411eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1544eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1412{ 1545{
1420 1553
1421#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1554#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1422#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1555#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1423 1556
1424static void 1557static void
1425eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1558eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1426{ 1559{
1427 unsigned char bits [9 + sizeof (ino_t) * 8]; 1560 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1428 unsigned char *bit = bits; 1561 unsigned char *bit = bits;
1429 1562
1430 assert (CHAR_BIT == 8); 1563 assert (CHAR_BIT == 8);
1431 assert (sizeof (eio_dirent) * 8 < 256); 1564 assert (sizeof (eio_dirent) * 8 < 256);
1432 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1565 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1434 1567
1435 if (size <= EIO_SORT_FAST) 1568 if (size <= EIO_SORT_FAST)
1436 return; 1569 return;
1437 1570
1438 /* first prepare an array of bits to test in our radix sort */ 1571 /* first prepare an array of bits to test in our radix sort */
1439 /* try to take endianness into account, as well as differences in ino_t sizes */ 1572 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1440 /* inode_bits must contain all inodes ORed together */ 1573 /* inode_bits must contain all inodes ORed together */
1441 /* which is used to skip bits that are 0 everywhere, which is very common */ 1574 /* which is used to skip bits that are 0 everywhere, which is very common */
1442 { 1575 {
1443 ino_t endianness; 1576 eio_ino_t endianness;
1444 int i, j; 1577 int i, j;
1445 1578
1446 /* we store the byte offset of byte n into byte n of "endianness" */ 1579 /* we store the byte offset of byte n into byte n of "endianness" */
1447 for (i = 0; i < sizeof (ino_t); ++i) 1580 for (i = 0; i < sizeof (eio_ino_t); ++i)
1448 ((unsigned char *)&endianness)[i] = i; 1581 ((unsigned char *)&endianness)[i] = i;
1449 1582
1450 *bit++ = 0; 1583 *bit++ = 0;
1451 1584
1452 for (i = 0; i < sizeof (ino_t); ++i) 1585 for (i = 0; i < sizeof (eio_ino_t); ++i)
1453 { 1586 {
1454 /* shifting off the byte offsets out of "endianness" */ 1587 /* shifting off the byte offsets out of "endianness" */
1455 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1588 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1456 endianness >>= 8; 1589 endianness >>= 8;
1457 1590
1458 for (j = 0; j < 8; ++j) 1591 for (j = 0; j < 8; ++j)
1459 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1592 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1460 *bit++ = offs + j; 1593 *bit++ = offs + j;
1461 } 1594 }
1462 1595
1463 for (j = 0; j < 8; ++j) 1596 for (j = 0; j < 8; ++j)
1464 if (score_bits & (1 << j)) 1597 if (score_bits & (1 << j))
1465 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1598 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1466 } 1599 }
1467 1600
1468 /* now actually do the sorting (a variant of MSD radix sort) */ 1601 /* now actually do the sorting (a variant of MSD radix sort) */
1469 { 1602 {
1470 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1603 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1471 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1604 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1472 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1605 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1473 int stk_idx = 0; 1606 int stk_idx = 0;
1474 1607
1475 base_stk [stk_idx] = dents; 1608 base_stk [stk_idx] = dents;
1476 end_stk [stk_idx] = dents + size; 1609 end_stk [stk_idx] = dents + size;
1477 bit_stk [stk_idx] = bit - 1; 1610 bit_stk [stk_idx] = bit - 1;
1556 } 1689 }
1557 } 1690 }
1558} 1691}
1559 1692
1560static void 1693static void
1561eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1694eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1562{ 1695{
1563 if (size <= 1) 1696 if (size <= 1)
1564 return; /* our insertion sort relies on size > 0 */ 1697 return; /* our insertion sort relies on size > 0 */
1565 1698
1566 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1699 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1574 1707
1575/* read a full directory */ 1708/* read a full directory */
1576static void 1709static void
1577eio__scandir (eio_req *req, etp_worker *self) 1710eio__scandir (eio_req *req, etp_worker *self)
1578{ 1711{
1579 DIR *dirp;
1580 EIO_STRUCT_DIRENT *entp;
1581 char *name, *names; 1712 char *name, *names;
1582 int namesalloc = 4096; 1713 int namesalloc = 4096 - sizeof (void *) * 4;
1583 int namesoffs = 0; 1714 int namesoffs = 0;
1584 int flags = req->int1; 1715 int flags = req->int1;
1585 eio_dirent *dents = 0; 1716 eio_dirent *dents = 0;
1586 int dentalloc = 128; 1717 int dentalloc = 128;
1587 int dentoffs = 0; 1718 int dentoffs = 0;
1588 ino_t inode_bits = 0; 1719 eio_ino_t inode_bits = 0;
1720#ifdef _WIN32
1721 HANDLE dirp;
1722 WIN32_FIND_DATA entp;
1723#else
1724 DIR *dirp;
1725 EIO_STRUCT_DIRENT *entp;
1726#endif
1589 1727
1590 req->result = -1; 1728 req->result = -1;
1591 1729
1592 if (!(flags & EIO_READDIR_DENTS)) 1730 if (!(flags & EIO_READDIR_DENTS))
1593 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1731 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1594 1732
1595 X_LOCK (wrklock); 1733#ifdef _WIN32
1596 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1734 {
1735 int len = strlen ((const char *)req->ptr1);
1736 char *path = malloc (MAX_PATH);
1737 const char *fmt;
1738 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1739
1740 if (!len)
1741 fmt = "./*";
1742 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1743 fmt = "%s*";
1744 else
1745 fmt = "%s/*";
1746
1747 _snprintf (path, MAX_PATH, fmt, reqpath);
1748 dirp = FindFirstFile (path, &entp);
1749 free (path);
1750
1751 if (dirp == INVALID_HANDLE_VALUE)
1752 {
1753 /* should steal _dosmaperr */
1754 switch (GetLastError ())
1755 {
1756 case ERROR_FILE_NOT_FOUND:
1757 req->result = 0;
1758 break;
1759
1760 case ERROR_INVALID_NAME:
1761 case ERROR_PATH_NOT_FOUND:
1762 case ERROR_NO_MORE_FILES:
1763 errno = ENOENT;
1764 break;
1765
1766 case ERROR_NOT_ENOUGH_MEMORY:
1767 errno = ENOMEM;
1768 break;
1769
1770 default:
1771 errno = EINVAL;
1772 break;
1773 }
1774
1775 return;
1776 }
1777 }
1778#else
1779 #if HAVE_AT
1780 if (req->wd)
1781 {
1782 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1783
1784 if (fd < 0)
1785 return;
1786
1787 dirp = fdopendir (fd);
1788
1789 if (!dirp)
1790 close (fd);
1791 }
1792 else
1597 self->dirp = dirp = opendir (req->ptr1); 1793 dirp = opendir (req->ptr1);
1794 #else
1795 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1796 #endif
1797
1798 if (!dirp)
1799 return;
1800#endif
1598 1801
1599 if (req->flags & EIO_FLAG_PTR1_FREE) 1802 if (req->flags & EIO_FLAG_PTR1_FREE)
1600 free (req->ptr1); 1803 free (req->ptr1);
1601 1804
1602 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1805 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1603 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1806 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1604 req->ptr2 = names = malloc (namesalloc); 1807 req->ptr2 = names = malloc (namesalloc);
1605 X_UNLOCK (wrklock);
1606 1808
1607 if (dirp && names && (!flags || dents)) 1809 if (!names || (flags && !dents))
1810 return;
1811
1608 for (;;) 1812 for (;;)
1609 { 1813 {
1814 int done;
1815
1816#ifdef _WIN32
1817 done = !dirp;
1818#else
1610 errno = 0; 1819 errno = 0;
1611 entp = readdir (dirp); 1820 entp = readdir (dirp);
1821 done = !entp;
1822#endif
1612 1823
1613 if (!entp) 1824 if (done)
1614 { 1825 {
1826#ifndef _WIN32
1827 int old_errno = errno;
1828 closedir (dirp);
1829 errno = old_errno;
1830
1615 if (errno) 1831 if (errno)
1616 break; 1832 break;
1833#endif
1617 1834
1618 /* sort etc. */ 1835 /* sort etc. */
1619 req->int1 = flags; 1836 req->int1 = flags;
1620 req->result = dentoffs; 1837 req->result = dentoffs;
1621 1838
1622 if (flags & EIO_READDIR_STAT_ORDER) 1839 if (flags & EIO_READDIR_STAT_ORDER)
1623 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1840 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1624 else if (flags & EIO_READDIR_DIRS_FIRST) 1841 else if (flags & EIO_READDIR_DIRS_FIRST)
1625 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1842 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1626 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1843 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1627 else 1844 else
1845 {
1846 /* in this case, all is known, and we just put dirs first and sort them */
1847 eio_dirent *oth = dents + dentoffs;
1848 eio_dirent *dir = dents;
1849
1850 /* now partition dirs to the front, and non-dirs to the back */
1851 /* by walking from both sides and swapping if necessary */
1852 while (oth > dir)
1853 {
1854 if (dir->type == EIO_DT_DIR)
1855 ++dir;
1856 else if ((--oth)->type == EIO_DT_DIR)
1857 {
1858 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1859
1860 ++dir;
1861 }
1862 }
1863
1864 /* now sort the dirs only (dirs all have the same score) */
1865 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1866 }
1867
1868 break;
1869 }
1870
1871 /* now add the entry to our list(s) */
1872 name = D_NAME (entp);
1873
1874 /* skip . and .. entries */
1875 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1876 {
1877 int len = D_NAMLEN (entp) + 1;
1878
1879 while (ecb_expect_false (namesoffs + len > namesalloc))
1880 {
1881 namesalloc *= 2;
1882 req->ptr2 = names = realloc (names, namesalloc);
1883
1884 if (!names)
1885 break;
1886 }
1887
1888 memcpy (names + namesoffs, name, len);
1889
1890 if (dents)
1891 {
1892 struct eio_dirent *ent;
1893
1894 if (ecb_expect_false (dentoffs == dentalloc))
1628 { 1895 {
1629 /* in this case, all is known, and we just put dirs first and sort them */ 1896 dentalloc *= 2;
1897 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1898
1899 if (!dents)
1900 break;
1901 }
1902
1630 eio_dirent *oth = dents + dentoffs; 1903 ent = dents + dentoffs;
1631 eio_dirent *dir = dents;
1632 1904
1633 /* now partition dirs to the front, and non-dirs to the back */ 1905 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1634 /* by walking from both sides and swapping if necessary */ 1906 ent->namelen = len - 1;
1635 while (oth > dir) 1907 ent->inode = D_INO (entp);
1908
1909 inode_bits |= ent->inode;
1910
1911 switch (D_TYPE (entp))
1912 {
1913 default:
1914 ent->type = EIO_DT_UNKNOWN;
1915 flags |= EIO_READDIR_FOUND_UNKNOWN;
1916 break;
1917
1918 #ifdef DT_FIFO
1919 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1920 #endif
1921 #ifdef DT_CHR
1922 case DT_CHR: ent->type = EIO_DT_CHR; break;
1923 #endif
1924 #ifdef DT_MPC
1925 case DT_MPC: ent->type = EIO_DT_MPC; break;
1926 #endif
1927 #ifdef DT_DIR
1928 case DT_DIR: ent->type = EIO_DT_DIR; break;
1929 #endif
1930 #ifdef DT_NAM
1931 case DT_NAM: ent->type = EIO_DT_NAM; break;
1932 #endif
1933 #ifdef DT_BLK
1934 case DT_BLK: ent->type = EIO_DT_BLK; break;
1935 #endif
1936 #ifdef DT_MPB
1937 case DT_MPB: ent->type = EIO_DT_MPB; break;
1938 #endif
1939 #ifdef DT_REG
1940 case DT_REG: ent->type = EIO_DT_REG; break;
1941 #endif
1942 #ifdef DT_NWK
1943 case DT_NWK: ent->type = EIO_DT_NWK; break;
1944 #endif
1945 #ifdef DT_CMP
1946 case DT_CMP: ent->type = EIO_DT_CMP; break;
1947 #endif
1948 #ifdef DT_LNK
1949 case DT_LNK: ent->type = EIO_DT_LNK; break;
1950 #endif
1951 #ifdef DT_SOCK
1952 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1953 #endif
1954 #ifdef DT_DOOR
1955 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1956 #endif
1957 #ifdef DT_WHT
1958 case DT_WHT: ent->type = EIO_DT_WHT; break;
1959 #endif
1960 }
1961
1962 ent->score = 7;
1963
1964 if (flags & EIO_READDIR_DIRS_FIRST)
1965 {
1966 if (ent->type == EIO_DT_UNKNOWN)
1636 { 1967 {
1637 if (dir->type == EIO_DT_DIR) 1968 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1638 ++dir; 1969 ent->score = 1;
1639 else if ((--oth)->type == EIO_DT_DIR) 1970 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1640 { 1971 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1641 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1642
1643 ++dir;
1644 }
1645 } 1972 }
1646 1973 else if (ent->type == EIO_DT_DIR)
1647 /* now sort the dirs only (dirs all have the same score) */ 1974 ent->score = 0;
1648 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1649 } 1975 }
1650
1651 break;
1652 }
1653
1654 /* now add the entry to our list(s) */
1655 name = entp->d_name;
1656
1657 /* skip . and .. entries */
1658 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1659 {
1660 int len = D_NAMLEN (entp) + 1;
1661
1662 while (ecb_expect_false (namesoffs + len > namesalloc))
1663 {
1664 namesalloc *= 2;
1665 X_LOCK (wrklock);
1666 req->ptr2 = names = realloc (names, namesalloc);
1667 X_UNLOCK (wrklock);
1668
1669 if (!names)
1670 break;
1671 } 1976 }
1672 1977
1673 memcpy (names + namesoffs, name, len);
1674
1675 if (dents)
1676 {
1677 struct eio_dirent *ent;
1678
1679 if (ecb_expect_false (dentoffs == dentalloc))
1680 {
1681 dentalloc *= 2;
1682 X_LOCK (wrklock);
1683 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1684 X_UNLOCK (wrklock);
1685
1686 if (!dents)
1687 break;
1688 }
1689
1690 ent = dents + dentoffs;
1691
1692 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1693 ent->namelen = len - 1;
1694 ent->inode = D_INO (entp);
1695
1696 inode_bits |= ent->inode;
1697
1698 switch (D_TYPE (entp))
1699 {
1700 default:
1701 ent->type = EIO_DT_UNKNOWN;
1702 flags |= EIO_READDIR_FOUND_UNKNOWN;
1703 break;
1704
1705 #ifdef DT_FIFO
1706 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1707 #endif
1708 #ifdef DT_CHR
1709 case DT_CHR: ent->type = EIO_DT_CHR; break;
1710 #endif
1711 #ifdef DT_MPC
1712 case DT_MPC: ent->type = EIO_DT_MPC; break;
1713 #endif
1714 #ifdef DT_DIR
1715 case DT_DIR: ent->type = EIO_DT_DIR; break;
1716 #endif
1717 #ifdef DT_NAM
1718 case DT_NAM: ent->type = EIO_DT_NAM; break;
1719 #endif
1720 #ifdef DT_BLK
1721 case DT_BLK: ent->type = EIO_DT_BLK; break;
1722 #endif
1723 #ifdef DT_MPB
1724 case DT_MPB: ent->type = EIO_DT_MPB; break;
1725 #endif
1726 #ifdef DT_REG
1727 case DT_REG: ent->type = EIO_DT_REG; break;
1728 #endif
1729 #ifdef DT_NWK
1730 case DT_NWK: ent->type = EIO_DT_NWK; break;
1731 #endif
1732 #ifdef DT_CMP
1733 case DT_CMP: ent->type = EIO_DT_CMP; break;
1734 #endif
1735 #ifdef DT_LNK
1736 case DT_LNK: ent->type = EIO_DT_LNK; break;
1737 #endif
1738 #ifdef DT_SOCK
1739 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1740 #endif
1741 #ifdef DT_DOOR
1742 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1743 #endif
1744 #ifdef DT_WHT
1745 case DT_WHT: ent->type = EIO_DT_WHT; break;
1746 #endif
1747 }
1748
1749 ent->score = 7;
1750
1751 if (flags & EIO_READDIR_DIRS_FIRST)
1752 {
1753 if (ent->type == EIO_DT_UNKNOWN)
1754 {
1755 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756 ent->score = 1;
1757 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1758 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759 }
1760 else if (ent->type == EIO_DT_DIR)
1761 ent->score = 0;
1762 }
1763 }
1764
1765 namesoffs += len; 1978 namesoffs += len;
1766 ++dentoffs; 1979 ++dentoffs;
1767 } 1980 }
1768 1981
1769 if (EIO_CANCELLED (req)) 1982 if (EIO_CANCELLED (req))
1770 { 1983 {
1771 errno = ECANCELED; 1984 errno = ECANCELED;
1772 break; 1985 break;
1773 } 1986 }
1987
1988#ifdef _WIN32
1989 if (!FindNextFile (dirp, &entp))
1990 {
1991 FindClose (dirp);
1992 dirp = 0;
1774 } 1993 }
1994#endif
1995 }
1775} 1996}
1997
1998/*****************************************************************************/
1999/* working directory stuff */
2000/* various deficiencies in the posix 2008 api force us to */
2001/* keep the absolute path in string form at all times */
2002/* fuck yeah. */
2003
2004#if !HAVE_AT
2005
2006/* a bit like realpath, but usually faster because it doesn'T have to return */
2007/* an absolute or canonical path */
2008static const char *
2009wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2010{
2011 if (!wd || *path == '/')
2012 return path;
2013
2014 if (path [0] == '.' && !path [1])
2015 return wd->str;
2016
2017 {
2018 int l1 = wd->len;
2019 int l2 = strlen (path);
2020
2021 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2022
2023 memcpy (res, wd->str, l1);
2024 res [l1] = '/';
2025 memcpy (res + l1 + 1, path, l2 + 1);
2026
2027 return res;
2028 }
2029}
2030
2031#endif
2032
2033static eio_wd
2034eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2035{
2036 int fd;
2037 eio_wd res;
2038 int len = eio__realpath (tmpbuf, wd, path);
2039
2040 if (len < 0)
2041 return EIO_INVALID_WD;
2042
2043#if HAVE_AT
2044 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2045
2046 if (fd < 0)
2047 return EIO_INVALID_WD;
2048#endif
2049
2050 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2051
2052#if HAVE_AT
2053 res->fd = fd;
2054#endif
2055
2056 res->len = len;
2057 memcpy (res->str, tmpbuf->ptr, len);
2058 res->str [len] = 0;
2059
2060 return res;
2061}
2062
2063eio_wd
2064eio_wd_open_sync (eio_wd wd, const char *path)
2065{
2066 struct tmpbuf tmpbuf = { 0 };
2067 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2068 free (tmpbuf.ptr);
2069
2070 return wd;
2071}
2072
2073void
2074eio_wd_close_sync (eio_wd wd)
2075{
2076 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2077 {
2078 #if HAVE_AT
2079 close (wd->fd);
2080 #endif
2081 free (wd);
2082 }
2083}
2084
2085#if HAVE_AT
2086
2087/* they forgot these */
2088
2089static int
2090eio__truncateat (int dirfd, const char *path, off_t length)
2091{
2092 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2093 int res;
2094
2095 if (fd < 0)
2096 return fd;
2097
2098 res = ftruncate (fd, length);
2099 close (fd);
2100 return res;
2101}
2102
2103static int
2104eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2105{
2106 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2107 int res;
2108
2109 if (fd < 0)
2110 return fd;
2111
2112 res = fstatvfs (fd, buf);
2113 close (fd);
2114 return res;
2115
2116}
2117
2118#endif
1776 2119
1777/*****************************************************************************/ 2120/*****************************************************************************/
1778 2121
1779#define ALLOC(len) \ 2122#define ALLOC(len) \
1780 if (!req->ptr2) \ 2123 if (!req->ptr2) \
1789 req->result = -1; \ 2132 req->result = -1; \
1790 break; \ 2133 break; \
1791 } \ 2134 } \
1792 } 2135 }
1793 2136
2137static void ecb_noinline ecb_cold
2138etp_proc_init (void)
2139{
2140#if HAVE_PRCTL_SET_NAME
2141 /* provide a more sensible "thread name" */
2142 char name[16 + 1];
2143 const int namelen = sizeof (name) - 1;
2144 int len;
2145
2146 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2147 name [namelen] = 0;
2148 len = strlen (name);
2149 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2150 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2151#endif
2152}
2153
1794X_THREAD_PROC (etp_proc) 2154X_THREAD_PROC (etp_proc)
1795{ 2155{
1796 ETP_REQ *req; 2156 ETP_REQ *req;
1797 struct timespec ts; 2157 struct timespec ts;
1798 etp_worker *self = (etp_worker *)thr_arg; 2158 etp_worker *self = (etp_worker *)thr_arg;
1799 int timeout; 2159
2160 etp_proc_init ();
1800 2161
1801 /* try to distribute timeouts somewhat evenly */ 2162 /* try to distribute timeouts somewhat evenly */
1802 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2163 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1803 2164
1804 for (;;) 2165 for (;;)
1807 2168
1808 X_LOCK (reqlock); 2169 X_LOCK (reqlock);
1809 2170
1810 for (;;) 2171 for (;;)
1811 { 2172 {
1812 self->req = req = reqq_shift (&req_queue); 2173 req = reqq_shift (&req_queue);
1813 2174
1814 if (req) 2175 if (req)
1815 break; 2176 break;
1816 2177
1817 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ 2178 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1855 ++npending; 2216 ++npending;
1856 2217
1857 if (!reqq_push (&res_queue, req) && want_poll_cb) 2218 if (!reqq_push (&res_queue, req) && want_poll_cb)
1858 want_poll_cb (); 2219 want_poll_cb ();
1859 2220
1860 self->req = 0;
1861 etp_worker_clear (self); 2221 etp_worker_clear (self);
1862 2222
1863 X_UNLOCK (reslock); 2223 X_UNLOCK (reslock);
1864 } 2224 }
1865 2225
1866quit: 2226quit:
2227 free (req);
2228
1867 X_LOCK (wrklock); 2229 X_LOCK (wrklock);
1868 etp_worker_free (self); 2230 etp_worker_free (self);
1869 X_UNLOCK (wrklock); 2231 X_UNLOCK (wrklock);
1870 2232
1871 return 0; 2233 return 0;
1883eio_api_destroy (eio_req *req) 2245eio_api_destroy (eio_req *req)
1884{ 2246{
1885 free (req); 2247 free (req);
1886} 2248}
1887 2249
1888#define REQ(rtype) \ 2250#define REQ(rtype) \
1889 eio_req *req; \ 2251 eio_req *req; \
1890 \ 2252 \
1891 req = (eio_req *)calloc (1, sizeof *req); \ 2253 req = (eio_req *)calloc (1, sizeof *req); \
1892 if (!req) \ 2254 if (!req) \
1893 return 0; \ 2255 return 0; \
1910 } 2272 }
1911 2273
1912static void 2274static void
1913eio_execute (etp_worker *self, eio_req *req) 2275eio_execute (etp_worker *self, eio_req *req)
1914{ 2276{
2277#if HAVE_AT
2278 int dirfd;
2279#else
2280 const char *path;
2281#endif
2282
1915 if (ecb_expect_false (EIO_CANCELLED (req))) 2283 if (ecb_expect_false (EIO_CANCELLED (req)))
1916 { 2284 {
1917 req->result = -1; 2285 req->result = -1;
1918 req->errorno = ECANCELED; 2286 req->errorno = ECANCELED;
1919 return; 2287 return;
1920 } 2288 }
1921 2289
2290 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2291 {
2292 req->result = -1;
2293 req->errorno = ENOENT;
2294 return;
2295 }
2296
2297 if (req->type >= EIO_OPEN)
2298 {
2299 #if HAVE_AT
2300 dirfd = WD2FD (req->wd);
2301 #else
2302 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2303 #endif
2304 }
2305
1922 switch (req->type) 2306 switch (req->type)
1923 { 2307 {
2308 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2309 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2310 break;
2311 case EIO_WD_CLOSE: req->result = 0;
2312 eio_wd_close_sync (req->wd); break;
2313
2314 case EIO_SEEK: eio__lseek (req); break;
1924 case EIO_READ: ALLOC (req->size); 2315 case EIO_READ: ALLOC (req->size);
1925 req->result = req->offs >= 0 2316 req->result = req->offs >= 0
1926 ? pread (req->int1, req->ptr2, req->size, req->offs) 2317 ? pread (req->int1, req->ptr2, req->size, req->offs)
1927 : read (req->int1, req->ptr2, req->size); break; 2318 : read (req->int1, req->ptr2, req->size); break;
1928 case EIO_WRITE: req->result = req->offs >= 0 2319 case EIO_WRITE: req->result = req->offs >= 0
1929 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2320 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1930 : write (req->int1, req->ptr2, req->size); break; 2321 : write (req->int1, req->ptr2, req->size); break;
1931 2322
1932 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2323 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1933 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2324 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2325
2326#if HAVE_AT
1934 2327
1935 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2328 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1936 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2329 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1937 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2330 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2331 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2332 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2333 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2334 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2335 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2336
2337 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2338 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2339 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2340 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2341 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2342 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2343 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2344 case EIO_READLINK: ALLOC (PATH_MAX);
2345 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2346 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2347 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2348 case EIO_UTIME:
2349 case EIO_FUTIME:
2350 {
2351 struct timespec ts[2];
2352 struct timespec *times;
2353
2354 if (req->nv1 != -1. || req->nv2 != -1.)
2355 {
2356 ts[0].tv_sec = req->nv1;
2357 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2358 ts[1].tv_sec = req->nv2;
2359 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2360
2361 times = ts;
2362 }
2363 else
2364 times = 0;
2365
2366 req->result = req->type == EIO_FUTIME
2367 ? futimens (req->int1, times)
2368 : utimensat (dirfd, req->ptr1, times, 0);
2369 }
2370 break;
2371
2372#else
2373
2374 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2375 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2376 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1938 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2377 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2378 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2379 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2380 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2381 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2382
2383 case EIO_UNLINK: req->result = unlink (path ); break;
2384 case EIO_RMDIR: req->result = rmdir (path ); break;
2385 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2386 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2387 case EIO_LINK: req->result = link (path , req->ptr2); break;
2388 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2389 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2390 case EIO_READLINK: ALLOC (PATH_MAX);
2391 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2392 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2393 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2394
2395 case EIO_UTIME:
2396 case EIO_FUTIME:
2397 {
2398 struct timeval tv[2];
2399 struct timeval *times;
2400
2401 if (req->nv1 != -1. || req->nv2 != -1.)
2402 {
2403 tv[0].tv_sec = req->nv1;
2404 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2405 tv[1].tv_sec = req->nv2;
2406 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2407
2408 times = tv;
2409 }
2410 else
2411 times = 0;
2412
2413 req->result = req->type == EIO_FUTIME
2414 ? futimes (req->int1, times)
2415 : utimes (req->ptr1, times);
2416 }
2417 break;
2418
2419#endif
2420
2421 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2422 {
2423 ALLOC (req->result);
2424 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2425 }
2426 break;
2427
1939 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2428 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1940 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2429 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1941 2430
1942 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1943 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1944 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2431 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1945 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2432 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1946 2433
1947 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1948 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2434 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1949 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2435 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1951 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1952 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2436 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1953 2437
1954 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1955 case EIO_CLOSE: req->result = close (req->int1); break; 2438 case EIO_CLOSE: req->result = close (req->int1); break;
1956 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2439 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1957 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1958 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1959 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1960 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1961 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1962 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1963 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1964
1965 case EIO_REALPATH: eio__realpath (req, self); break;
1966
1967 case EIO_READLINK: ALLOC (PATH_MAX);
1968 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1969
1970 case EIO_SYNC: req->result = 0; sync (); break; 2440 case EIO_SYNC: req->result = 0; sync (); break;
1971 case EIO_FSYNC: req->result = fsync (req->int1); break; 2441 case EIO_FSYNC: req->result = fsync (req->int1); break;
1972 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2442 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2443 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2444 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1973 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2445 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1974 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2446 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1975 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2447 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1976 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2448 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1977 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1978 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2449 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1979 2450
1980 case EIO_READDIR: eio__scandir (req, self); break; 2451 case EIO_READDIR: eio__scandir (req, self); break;
1981 2452
1982 case EIO_BUSY: 2453 case EIO_BUSY:
1992 req->result = select (0, 0, 0, 0, &tv); 2463 req->result = select (0, 0, 0, 0, &tv);
1993 } 2464 }
1994#endif 2465#endif
1995 break; 2466 break;
1996 2467
1997 case EIO_UTIME:
1998 case EIO_FUTIME:
1999 {
2000 struct timeval tv[2];
2001 struct timeval *times;
2002
2003 if (req->nv1 != -1. || req->nv2 != -1.)
2004 {
2005 tv[0].tv_sec = req->nv1;
2006 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2007 tv[1].tv_sec = req->nv2;
2008 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2009
2010 times = tv;
2011 }
2012 else
2013 times = 0;
2014
2015 req->result = req->type == EIO_FUTIME
2016 ? futimes (req->int1, times)
2017 : utimes (req->ptr1, times);
2018 }
2019 break;
2020
2021 case EIO_GROUP: 2468 case EIO_GROUP:
2022 abort (); /* handled in eio_request */ 2469 abort (); /* handled in eio_request */
2023 2470
2024 case EIO_NOP: 2471 case EIO_NOP:
2025 req->result = 0; 2472 req->result = 0;
2028 case EIO_CUSTOM: 2475 case EIO_CUSTOM:
2029 req->feed (req); 2476 req->feed (req);
2030 break; 2477 break;
2031 2478
2032 default: 2479 default:
2033 errno = ENOSYS;
2034 req->result = -1; 2480 req->result = EIO_ENOSYS ();
2035 break; 2481 break;
2036 } 2482 }
2037 2483
2038 req->errorno = errno; 2484 req->errorno = errno;
2039} 2485}
2040 2486
2041#ifndef EIO_NO_WRAPPERS 2487#ifndef EIO_NO_WRAPPERS
2042 2488
2489eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_WD_OPEN); PATH; SEND;
2492}
2493
2494eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2497}
2498
2043eio_req *eio_nop (int pri, eio_cb cb, void *data) 2499eio_req *eio_nop (int pri, eio_cb cb, void *data)
2044{ 2500{
2045 REQ (EIO_NOP); SEND; 2501 REQ (EIO_NOP); SEND;
2046} 2502}
2047 2503
2063eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2519eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2064{ 2520{
2065 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2521 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2066} 2522}
2067 2523
2524eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2525{
2526 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2527}
2528
2529eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2530{
2531 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2532}
2533
2534eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2537}
2538
2068eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2539eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2069{ 2540{
2070 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2541 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2071} 2542}
2072 2543
2078eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2549eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2079{ 2550{
2080 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2551 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2081} 2552}
2082 2553
2083eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2084{
2085 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2086}
2087
2088eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2554eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2089{ 2555{
2090 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2556 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2091} 2557}
2092 2558
2093eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2094{
2095 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2096}
2097
2098eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2559eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2099{ 2560{
2100 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2561 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2101} 2562}
2102 2563
2103eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2564eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2104{ 2565{
2105 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2566 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2106} 2567}
2107 2568
2569eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2572}
2573
2108eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2574eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2109{ 2575{
2110 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2576 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2111} 2577}
2112 2578
2138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2604eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2139{ 2605{
2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2606 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2141} 2607}
2142 2608
2143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2609eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2144{ 2610{
2145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2611 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2146} 2612}
2147 2613
2148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2614eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2634eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2169{ 2635{
2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2636 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2171} 2637}
2172 2638
2173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2639eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2174{ 2640{
2175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2641 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2176} 2642}
2177 2643
2178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2644eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2325} 2791}
2326 2792
2327/*****************************************************************************/ 2793/*****************************************************************************/
2328/* misc garbage */ 2794/* misc garbage */
2329 2795
2330ssize_t 2796eio_ssize_t
2331eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2797eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2332{ 2798{
2333 etp_worker wrk;
2334 ssize_t ret;
2335
2336 wrk.dbuf = 0;
2337
2338 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2799 return eio__sendfile (ofd, ifd, offset, count);
2339
2340 if (wrk.dbuf)
2341 free (wrk.dbuf);
2342
2343 return ret;
2344} 2800}
2345 2801

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines