2 * Server-side file descriptor management
4 * Copyright (C) 2000, 2003 Alexandre Julliard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
23 #include "wine/port.h"
38 #ifdef HAVE_SYS_POLL_H
41 #ifdef HAVE_LINUX_MAJOR_H
42 #include <linux/major.h>
44 #ifdef HAVE_SYS_STATVFS_H
45 #include <sys/statvfs.h>
48 /* Work around a conflict with Solaris' system list defined in sys/list.h. */
50 #define list_next SYSLIST_NEXT
51 #define list_prev SYSLIST_PREV
52 #define list_head SYSLIST_HEAD
53 #define list_tail SYSLIST_TAIL
54 #define list_move_tail SYSLIST_MOVE_TAIL
55 #define list_remove SYSLIST_REMOVE
65 #ifdef HAVE_SYS_PARAM_H
66 #include <sys/param.h>
68 #ifdef HAVE_SYS_MOUNT_H
69 #include <sys/mount.h>
71 #ifdef HAVE_SYS_STATFS_H
72 #include <sys/statfs.h>
74 #ifdef HAVE_SYS_SYSCTL_H
75 #include <sys/sysctl.h>
77 #ifdef HAVE_SYS_EVENT_H
78 #include <sys/event.h>
88 #include <sys/mkdev.h>
89 #elif defined(MAJOR_IN_SYSMACROS)
90 #include <sys/sysmacros.h>
92 #include <sys/types.h>
94 #ifdef HAVE_SYS_SYSCALL_H
95 #include <sys/syscall.h>
99 #define WIN32_NO_STATUS
106 #include "winternl.h"
107 #include "winioctl.h"
110 #if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
111 # include <sys/epoll.h>
113 #elif defined(linux) && defined(__i386__) && defined(HAVE_STDINT_H)
115 # define EPOLLIN POLLIN
116 # define EPOLLOUT POLLOUT
117 # define EPOLLERR POLLERR
118 # define EPOLLHUP POLLHUP
119 # define EPOLL_CTL_ADD 1
120 # define EPOLL_CTL_DEL 2
121 # define EPOLL_CTL_MOD 3
123 typedef union epoll_data
137 static inline int epoll_create( int size
)
139 return syscall( 254 /*NR_epoll_create*/, size
);
142 static inline int epoll_ctl( int epfd
, int op
, int fd
, const struct epoll_event
*event
)
144 return syscall( 255 /*NR_epoll_ctl*/, epfd
, op
, fd
, event
);
147 static inline int epoll_wait( int epfd
, struct epoll_event
*events
, int maxevents
, int timeout
)
149 return syscall( 256 /*NR_epoll_wait*/, epfd
, events
, maxevents
, timeout
);
152 #endif /* linux && __i386__ && HAVE_STDINT_H */
154 #if defined(HAVE_PORT_H) && defined(HAVE_PORT_CREATE)
156 # define USE_EVENT_PORTS
157 #endif /* HAVE_PORT_H && HAVE_PORT_CREATE */
159 /* Because of the stupid Posix locking semantics, we need to keep
160 * track of all file descriptors referencing a given file, and not
161 * close a single one until all the locks are gone (sigh).
164 /* file descriptor object */
166 /* closed_fd is used to keep track of the unix fd belonging to a closed fd object */
169 struct list entry
; /* entry in inode closed list */
170 int unix_fd
; /* the unix file descriptor */
171 int unlink
; /* whether to unlink on close: -1 - implicit FILE_DELETE_ON_CLOSE, 1 - explicit disposition */
172 char *unix_name
; /* name to unlink on close, points to parent fd unix_name */
177 struct object obj
; /* object header */
178 const struct fd_ops
*fd_ops
; /* file descriptor operations */
179 struct inode
*inode
; /* inode that this fd belongs to */
180 struct list inode_entry
; /* entry in inode fd list */
181 struct closed_fd
*closed
; /* structure to store the unix fd at destroy time */
182 struct object
*user
; /* object using this file descriptor */
183 struct list locks
; /* list of locks on this fd */
184 unsigned int access
; /* file access (FILE_READ_DATA etc.) */
185 unsigned int options
; /* file options (FILE_DELETE_ON_CLOSE, FILE_SYNCHRONOUS...) */
186 unsigned int sharing
; /* file sharing mode */
187 char *unix_name
; /* unix file name */
188 WCHAR
*nt_name
; /* NT file name */
189 data_size_t nt_namelen
; /* length of NT file name */
190 int unix_fd
; /* unix file descriptor */
191 unsigned int no_fd_status
;/* status to return when unix_fd is -1 */
192 unsigned int cacheable
:1;/* can the fd be cached on the client side? */
193 unsigned int signaled
:1; /* is the fd signaled? */
194 unsigned int fs_locks
:1; /* can we use filesystem locks for this fd? */
195 int poll_index
; /* index of fd in poll array */
196 struct async_queue read_q
; /* async readers of this fd */
197 struct async_queue write_q
; /* async writers of this fd */
198 struct async_queue wait_q
; /* other async waiters of this fd */
199 struct completion
*completion
; /* completion object attached to this fd */
200 apc_param_t comp_key
; /* completion key to set in completion events */
201 unsigned int comp_flags
; /* completion flags */
204 static void fd_dump( struct object
*obj
, int verbose
);
205 static void fd_destroy( struct object
*obj
);
207 static const struct object_ops fd_ops
=
209 sizeof(struct fd
), /* size */
212 no_add_queue
, /* add_queue */
213 NULL
, /* remove_queue */
215 NULL
, /* satisfied */
216 no_signal
, /* signal */
217 no_get_fd
, /* get_fd */
218 default_map_access
, /* map_access */
219 default_get_sd
, /* get_sd */
220 default_set_sd
, /* set_sd */
221 no_get_full_name
, /* get_full_name */
222 no_lookup_name
, /* lookup_name */
223 no_link_name
, /* link_name */
224 NULL
, /* unlink_name */
225 no_open_file
, /* open_file */
226 no_kernel_obj_list
, /* get_kernel_obj_list */
227 no_close_handle
, /* close_handle */
228 fd_destroy
/* destroy */
233 #define DEVICE_HASH_SIZE 7
234 #define INODE_HASH_SIZE 17
238 struct object obj
; /* object header */
239 struct list entry
; /* entry in device hash list */
240 dev_t dev
; /* device number */
241 int removable
; /* removable device? (or -1 if unknown) */
242 struct list inode_hash
[INODE_HASH_SIZE
]; /* inodes hash table */
245 static void device_dump( struct object
*obj
, int verbose
);
246 static void device_destroy( struct object
*obj
);
248 static const struct object_ops device_ops
=
250 sizeof(struct device
), /* size */
252 device_dump
, /* dump */
253 no_add_queue
, /* add_queue */
254 NULL
, /* remove_queue */
256 NULL
, /* satisfied */
257 no_signal
, /* signal */
258 no_get_fd
, /* get_fd */
259 default_map_access
, /* map_access */
260 default_get_sd
, /* get_sd */
261 default_set_sd
, /* set_sd */
262 no_get_full_name
, /* get_full_name */
263 no_lookup_name
, /* lookup_name */
264 no_link_name
, /* link_name */
265 NULL
, /* unlink_name */
266 no_open_file
, /* open_file */
267 no_kernel_obj_list
, /* get_kernel_obj_list */
268 no_close_handle
, /* close_handle */
269 device_destroy
/* destroy */
276 struct object obj
; /* object header */
277 struct list entry
; /* inode hash list entry */
278 struct device
*device
; /* device containing this inode */
279 ino_t ino
; /* inode number */
280 struct list open
; /* list of open file descriptors */
281 struct list locks
; /* list of file locks */
282 struct list closed
; /* list of file descriptors to close at destroy time */
285 static void inode_dump( struct object
*obj
, int verbose
);
286 static void inode_destroy( struct object
*obj
);
288 static const struct object_ops inode_ops
=
290 sizeof(struct inode
), /* size */
292 inode_dump
, /* dump */
293 no_add_queue
, /* add_queue */
294 NULL
, /* remove_queue */
296 NULL
, /* satisfied */
297 no_signal
, /* signal */
298 no_get_fd
, /* get_fd */
299 default_map_access
, /* map_access */
300 default_get_sd
, /* get_sd */
301 default_set_sd
, /* set_sd */
302 no_get_full_name
, /* get_full_name */
303 no_lookup_name
, /* lookup_name */
304 no_link_name
, /* link_name */
305 NULL
, /* unlink_name */
306 no_open_file
, /* open_file */
307 no_kernel_obj_list
, /* get_kernel_obj_list */
308 no_close_handle
, /* close_handle */
309 inode_destroy
/* destroy */
312 /* file lock object */
316 struct object obj
; /* object header */
317 struct fd
*fd
; /* fd owning this lock */
318 struct list fd_entry
; /* entry in list of locks on a given fd */
319 struct list inode_entry
; /* entry in inode list of locks */
320 int shared
; /* shared lock? */
321 file_pos_t start
; /* locked region is interval [start;end) */
323 struct process
*process
; /* process owning this lock */
324 struct list proc_entry
; /* entry in list of locks owned by the process */
327 static void file_lock_dump( struct object
*obj
, int verbose
);
328 static int file_lock_signaled( struct object
*obj
, struct wait_queue_entry
*entry
);
330 static const struct object_ops file_lock_ops
=
332 sizeof(struct file_lock
), /* size */
334 file_lock_dump
, /* dump */
335 add_queue
, /* add_queue */
336 remove_queue
, /* remove_queue */
337 file_lock_signaled
, /* signaled */
338 no_satisfied
, /* satisfied */
339 no_signal
, /* signal */
340 no_get_fd
, /* get_fd */
341 default_map_access
, /* map_access */
342 default_get_sd
, /* get_sd */
343 default_set_sd
, /* set_sd */
344 no_get_full_name
, /* get_full_name */
345 no_lookup_name
, /* lookup_name */
346 no_link_name
, /* link_name */
347 NULL
, /* unlink_name */
348 no_open_file
, /* open_file */
349 no_kernel_obj_list
, /* get_kernel_obj_list */
350 no_close_handle
, /* close_handle */
351 no_destroy
/* destroy */
355 #define OFF_T_MAX (~((file_pos_t)1 << (8*sizeof(off_t)-1)))
356 #define FILE_POS_T_MAX (~(file_pos_t)0)
358 static file_pos_t max_unix_offset
= OFF_T_MAX
;
360 #define DUMP_LONG_LONG(val) do { \
361 if (sizeof(val) > sizeof(unsigned long) && (val) > ~0UL) \
362 fprintf( stderr, "%lx%08lx", (unsigned long)((unsigned long long)(val) >> 32), (unsigned long)(val) ); \
364 fprintf( stderr, "%lx", (unsigned long)(val) ); \
369 /****************************************************************/
370 /* timeouts support */
374 struct list entry
; /* entry in sorted timeout list */
375 abstime_t when
; /* timeout expiry */
376 timeout_callback callback
; /* callback function */
377 void *private; /* callback private data */
380 static struct list abs_timeout_list
= LIST_INIT(abs_timeout_list
); /* sorted absolute timeouts list */
381 static struct list rel_timeout_list
= LIST_INIT(rel_timeout_list
); /* sorted relative timeouts list */
382 timeout_t current_time
;
383 timeout_t monotonic_time
;
385 struct _KUSER_SHARED_DATA
*user_shared_data
= NULL
;
386 static const int user_shared_data_timeout
= 16;
388 static void set_user_shared_data_time(void)
390 timeout_t tick_count
= monotonic_time
/ 10000;
392 /* on X86 there should be total store order guarantees, so volatile is enough
393 * to ensure the stores aren't reordered by the compiler, and then they will
394 * always be seen in-order from other CPUs. On other archs, we need atomic
395 * intrinsics to guarantee that. */
396 #if defined(__i386__) || defined(__x86_64__)
397 user_shared_data
->SystemTime
.High2Time
= current_time
>> 32;
398 user_shared_data
->SystemTime
.LowPart
= current_time
;
399 user_shared_data
->SystemTime
.High1Time
= current_time
>> 32;
401 user_shared_data
->InterruptTime
.High2Time
= monotonic_time
>> 32;
402 user_shared_data
->InterruptTime
.LowPart
= monotonic_time
;
403 user_shared_data
->InterruptTime
.High1Time
= monotonic_time
>> 32;
405 user_shared_data
->TickCount
.High2Time
= tick_count
>> 32;
406 user_shared_data
->TickCount
.LowPart
= tick_count
;
407 user_shared_data
->TickCount
.High1Time
= tick_count
>> 32;
408 *(volatile ULONG
*)&user_shared_data
->TickCountLowDeprecated
= tick_count
;
410 __atomic_store_n(&user_shared_data
->SystemTime
.High2Time
, current_time
>> 32, __ATOMIC_SEQ_CST
);
411 __atomic_store_n(&user_shared_data
->SystemTime
.LowPart
, current_time
, __ATOMIC_SEQ_CST
);
412 __atomic_store_n(&user_shared_data
->SystemTime
.High1Time
, current_time
>> 32, __ATOMIC_SEQ_CST
);
414 __atomic_store_n(&user_shared_data
->InterruptTime
.High2Time
, monotonic_time
>> 32, __ATOMIC_SEQ_CST
);
415 __atomic_store_n(&user_shared_data
->InterruptTime
.LowPart
, monotonic_time
, __ATOMIC_SEQ_CST
);
416 __atomic_store_n(&user_shared_data
->InterruptTime
.High1Time
, monotonic_time
>> 32, __ATOMIC_SEQ_CST
);
418 __atomic_store_n(&user_shared_data
->TickCount
.High2Time
, tick_count
>> 32, __ATOMIC_SEQ_CST
);
419 __atomic_store_n(&user_shared_data
->TickCount
.LowPart
, tick_count
, __ATOMIC_SEQ_CST
);
420 __atomic_store_n(&user_shared_data
->TickCount
.High1Time
, tick_count
>> 32, __ATOMIC_SEQ_CST
);
421 __atomic_store_n(&user_shared_data
->TickCountLowDeprecated
, tick_count
, __ATOMIC_SEQ_CST
);
425 void set_current_time(void)
427 static const timeout_t ticks_1601_to_1970
= (timeout_t
)86400 * (369 * 365 + 89) * TICKS_PER_SEC
;
429 gettimeofday( &now
, NULL
);
430 current_time
= (timeout_t
)now
.tv_sec
* TICKS_PER_SEC
+ now
.tv_usec
* 10 + ticks_1601_to_1970
;
431 monotonic_time
= monotonic_counter();
432 if (user_shared_data
) set_user_shared_data_time();
435 /* add a timeout user */
436 struct timeout_user
*add_timeout_user( timeout_t when
, timeout_callback func
, void *private )
438 struct timeout_user
*user
;
441 if (!(user
= mem_alloc( sizeof(*user
) ))) return NULL
;
442 user
->when
= timeout_to_abstime( when
);
443 user
->callback
= func
;
444 user
->private = private;
446 /* Now insert it in the linked list */
450 LIST_FOR_EACH( ptr
, &abs_timeout_list
)
452 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
453 if (timeout
->when
>= user
->when
) break;
458 LIST_FOR_EACH( ptr
, &rel_timeout_list
)
460 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
461 if (timeout
->when
<= user
->when
) break;
464 list_add_before( ptr
, &user
->entry
);
468 /* remove a timeout user */
469 void remove_timeout_user( struct timeout_user
*user
)
471 list_remove( &user
->entry
);
475 /* return a text description of a timeout for debugging purposes */
476 const char *get_timeout_str( timeout_t timeout
)
478 static char buffer
[64];
481 if (!timeout
) return "0";
482 if (timeout
== TIMEOUT_INFINITE
) return "infinite";
484 if (timeout
< 0) /* relative */
486 secs
= -timeout
/ TICKS_PER_SEC
;
487 nsecs
= -timeout
% TICKS_PER_SEC
;
488 sprintf( buffer
, "+%ld.%07ld", secs
, nsecs
);
492 secs
= (timeout
- current_time
) / TICKS_PER_SEC
;
493 nsecs
= (timeout
- current_time
) % TICKS_PER_SEC
;
496 nsecs
+= TICKS_PER_SEC
;
500 sprintf( buffer
, "%x%08x (+%ld.%07ld)",
501 (unsigned int)(timeout
>> 32), (unsigned int)timeout
, secs
, nsecs
);
503 sprintf( buffer
, "%x%08x (-%ld.%07ld)",
504 (unsigned int)(timeout
>> 32), (unsigned int)timeout
,
505 -(secs
+ 1), TICKS_PER_SEC
- nsecs
);
511 /****************************************************************/
514 static struct fd
**poll_users
; /* users array */
515 static struct pollfd
*pollfd
; /* poll fd array */
516 static int nb_users
; /* count of array entries actually in use */
517 static int active_users
; /* current number of active users */
518 static int allocated_users
; /* count of allocated entries in the array */
519 static struct fd
**freelist
; /* list of free entries in the array */
521 static int get_next_timeout(void);
523 static inline void fd_poll_event( struct fd
*fd
, int event
)
525 fd
->fd_ops
->poll_event( fd
, event
);
530 static int epoll_fd
= -1;
532 static inline void init_epoll(void)
534 epoll_fd
= epoll_create( 128 );
537 /* set the events that epoll waits for on this fd; helper for set_fd_events */
538 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
)
540 struct epoll_event ev
;
543 if (epoll_fd
== -1) return;
545 if (events
== -1) /* stop waiting on this fd completely */
547 if (pollfd
[user
].fd
== -1) return; /* already removed */
550 else if (pollfd
[user
].fd
== -1)
552 if (pollfd
[user
].events
) return; /* stopped waiting on it, don't restart */
557 if (pollfd
[user
].events
== events
) return; /* nothing to do */
562 memset(&ev
.data
, 0, sizeof(ev
.data
));
565 if (epoll_ctl( epoll_fd
, ctl
, fd
->unix_fd
, &ev
) == -1)
567 if (errno
== ENOMEM
) /* not enough memory, give up on epoll */
572 else perror( "epoll_ctl" ); /* should not happen */
576 static inline void remove_epoll_user( struct fd
*fd
, int user
)
578 if (epoll_fd
== -1) return;
580 if (pollfd
[user
].fd
!= -1)
582 struct epoll_event dummy
;
583 epoll_ctl( epoll_fd
, EPOLL_CTL_DEL
, fd
->unix_fd
, &dummy
);
587 static inline void main_loop_epoll(void)
590 struct epoll_event events
[128];
592 assert( POLLIN
== EPOLLIN
);
593 assert( POLLOUT
== EPOLLOUT
);
594 assert( POLLERR
== EPOLLERR
);
595 assert( POLLHUP
== EPOLLHUP
);
597 if (epoll_fd
== -1) return;
601 timeout
= get_next_timeout();
603 if (!active_users
) break; /* last user removed by a timeout */
604 if (epoll_fd
== -1) break; /* an error occurred with epoll */
606 ret
= epoll_wait( epoll_fd
, events
, ARRAY_SIZE( events
), timeout
);
609 /* put the events into the pollfd array first, like poll does */
610 for (i
= 0; i
< ret
; i
++)
612 int user
= events
[i
].data
.u32
;
613 pollfd
[user
].revents
= events
[i
].events
;
616 /* read events from the pollfd array, as set_fd_events may modify them */
617 for (i
= 0; i
< ret
; i
++)
619 int user
= events
[i
].data
.u32
;
620 if (pollfd
[user
].revents
) fd_poll_event( poll_users
[user
], pollfd
[user
].revents
);
625 #elif defined(HAVE_KQUEUE)
627 static int kqueue_fd
= -1;
629 static inline void init_epoll(void)
631 #ifdef __APPLE__ /* kqueue support is broken in Mac OS < 10.5 */
634 size_t len
= sizeof(release
);
637 mib
[1] = KERN_OSRELEASE
;
638 if (sysctl( mib
, 2, release
, &len
, NULL
, 0 ) == -1) return;
639 if (atoi(release
) < 9) return;
641 kqueue_fd
= kqueue();
644 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
)
648 if (kqueue_fd
== -1) return;
650 EV_SET( &ev
[0], fd
->unix_fd
, EVFILT_READ
, 0, NOTE_LOWAT
, 1, (void *)(long)user
);
651 EV_SET( &ev
[1], fd
->unix_fd
, EVFILT_WRITE
, 0, NOTE_LOWAT
, 1, (void *)(long)user
);
653 if (events
== -1) /* stop waiting on this fd completely */
655 if (pollfd
[user
].fd
== -1) return; /* already removed */
656 ev
[0].flags
|= EV_DELETE
;
657 ev
[1].flags
|= EV_DELETE
;
659 else if (pollfd
[user
].fd
== -1)
661 if (pollfd
[user
].events
) return; /* stopped waiting on it, don't restart */
662 ev
[0].flags
|= EV_ADD
| ((events
& POLLIN
) ? EV_ENABLE
: EV_DISABLE
);
663 ev
[1].flags
|= EV_ADD
| ((events
& POLLOUT
) ? EV_ENABLE
: EV_DISABLE
);
667 if (pollfd
[user
].events
== events
) return; /* nothing to do */
668 ev
[0].flags
|= (events
& POLLIN
) ? EV_ENABLE
: EV_DISABLE
;
669 ev
[1].flags
|= (events
& POLLOUT
) ? EV_ENABLE
: EV_DISABLE
;
672 if (kevent( kqueue_fd
, ev
, 2, NULL
, 0, NULL
) == -1)
674 if (errno
== ENOMEM
) /* not enough memory, give up on kqueue */
679 else perror( "kevent" ); /* should not happen */
683 static inline void remove_epoll_user( struct fd
*fd
, int user
)
685 if (kqueue_fd
== -1) return;
687 if (pollfd
[user
].fd
!= -1)
691 EV_SET( &ev
[0], fd
->unix_fd
, EVFILT_READ
, EV_DELETE
, 0, 0, 0 );
692 EV_SET( &ev
[1], fd
->unix_fd
, EVFILT_WRITE
, EV_DELETE
, 0, 0, 0 );
693 kevent( kqueue_fd
, ev
, 2, NULL
, 0, NULL
);
697 static inline void main_loop_epoll(void)
700 struct kevent events
[128];
702 if (kqueue_fd
== -1) return;
706 timeout
= get_next_timeout();
708 if (!active_users
) break; /* last user removed by a timeout */
709 if (kqueue_fd
== -1) break; /* an error occurred with kqueue */
715 ts
.tv_sec
= timeout
/ 1000;
716 ts
.tv_nsec
= (timeout
% 1000) * 1000000;
717 ret
= kevent( kqueue_fd
, NULL
, 0, events
, ARRAY_SIZE( events
), &ts
);
719 else ret
= kevent( kqueue_fd
, NULL
, 0, events
, ARRAY_SIZE( events
), NULL
);
723 /* put the events into the pollfd array first, like poll does */
724 for (i
= 0; i
< ret
; i
++)
726 long user
= (long)events
[i
].udata
;
727 pollfd
[user
].revents
= 0;
729 for (i
= 0; i
< ret
; i
++)
731 long user
= (long)events
[i
].udata
;
732 if (events
[i
].filter
== EVFILT_READ
) pollfd
[user
].revents
|= POLLIN
;
733 else if (events
[i
].filter
== EVFILT_WRITE
) pollfd
[user
].revents
|= POLLOUT
;
734 if (events
[i
].flags
& EV_EOF
) pollfd
[user
].revents
|= POLLHUP
;
735 if (events
[i
].flags
& EV_ERROR
) pollfd
[user
].revents
|= POLLERR
;
738 /* read events from the pollfd array, as set_fd_events may modify them */
739 for (i
= 0; i
< ret
; i
++)
741 long user
= (long)events
[i
].udata
;
742 if (pollfd
[user
].revents
) fd_poll_event( poll_users
[user
], pollfd
[user
].revents
);
743 pollfd
[user
].revents
= 0;
748 #elif defined(USE_EVENT_PORTS)
750 static int port_fd
= -1;
752 static inline void init_epoll(void)
754 port_fd
= port_create();
757 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
)
761 if (port_fd
== -1) return;
763 if (events
== -1) /* stop waiting on this fd completely */
765 if (pollfd
[user
].fd
== -1) return; /* already removed */
766 port_dissociate( port_fd
, PORT_SOURCE_FD
, fd
->unix_fd
);
768 else if (pollfd
[user
].fd
== -1)
770 if (pollfd
[user
].events
) return; /* stopped waiting on it, don't restart */
771 ret
= port_associate( port_fd
, PORT_SOURCE_FD
, fd
->unix_fd
, events
, (void *)user
);
775 if (pollfd
[user
].events
== events
) return; /* nothing to do */
776 ret
= port_associate( port_fd
, PORT_SOURCE_FD
, fd
->unix_fd
, events
, (void *)user
);
781 if (errno
== ENOMEM
) /* not enough memory, give up on port_associate */
786 else perror( "port_associate" ); /* should not happen */
790 static inline void remove_epoll_user( struct fd
*fd
, int user
)
792 if (port_fd
== -1) return;
794 if (pollfd
[user
].fd
!= -1)
796 port_dissociate( port_fd
, PORT_SOURCE_FD
, fd
->unix_fd
);
800 static inline void main_loop_epoll(void)
802 int i
, nget
, ret
, timeout
;
803 port_event_t events
[128];
805 if (port_fd
== -1) return;
809 timeout
= get_next_timeout();
812 if (!active_users
) break; /* last user removed by a timeout */
813 if (port_fd
== -1) break; /* an error occurred with event completion */
819 ts
.tv_sec
= timeout
/ 1000;
820 ts
.tv_nsec
= (timeout
% 1000) * 1000000;
821 ret
= port_getn( port_fd
, events
, ARRAY_SIZE( events
), &nget
, &ts
);
823 else ret
= port_getn( port_fd
, events
, ARRAY_SIZE( events
), &nget
, NULL
);
825 if (ret
== -1) break; /* an error occurred with event completion */
829 /* put the events into the pollfd array first, like poll does */
830 for (i
= 0; i
< nget
; i
++)
832 long user
= (long)events
[i
].portev_user
;
833 pollfd
[user
].revents
= events
[i
].portev_events
;
836 /* read events from the pollfd array, as set_fd_events may modify them */
837 for (i
= 0; i
< nget
; i
++)
839 long user
= (long)events
[i
].portev_user
;
840 if (pollfd
[user
].revents
) fd_poll_event( poll_users
[user
], pollfd
[user
].revents
);
841 /* if we are still interested, reassociate the fd */
842 if (pollfd
[user
].fd
!= -1) {
843 port_associate( port_fd
, PORT_SOURCE_FD
, pollfd
[user
].fd
, pollfd
[user
].events
, (void *)user
);
849 #else /* HAVE_KQUEUE */
851 static inline void init_epoll(void) { }
852 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
) { }
853 static inline void remove_epoll_user( struct fd
*fd
, int user
) { }
854 static inline void main_loop_epoll(void) { }
856 #endif /* USE_EPOLL */
859 /* add a user in the poll array and return its index, or -1 on failure */
860 static int add_poll_user( struct fd
*fd
)
865 ret
= freelist
- poll_users
;
866 freelist
= (struct fd
**)poll_users
[ret
];
870 if (nb_users
== allocated_users
)
872 struct fd
**newusers
;
873 struct pollfd
*newpoll
;
874 int new_count
= allocated_users
? (allocated_users
+ allocated_users
/ 2) : 16;
875 if (!(newusers
= realloc( poll_users
, new_count
* sizeof(*poll_users
) ))) return -1;
876 if (!(newpoll
= realloc( pollfd
, new_count
* sizeof(*pollfd
) )))
879 poll_users
= newusers
;
884 poll_users
= newusers
;
886 if (!allocated_users
) init_epoll();
887 allocated_users
= new_count
;
892 pollfd
[ret
].events
= 0;
893 pollfd
[ret
].revents
= 0;
894 poll_users
[ret
] = fd
;
899 /* remove a user from the poll list */
900 static void remove_poll_user( struct fd
*fd
, int user
)
903 assert( poll_users
[user
] == fd
);
905 remove_epoll_user( fd
, user
);
906 pollfd
[user
].fd
= -1;
907 pollfd
[user
].events
= 0;
908 pollfd
[user
].revents
= 0;
909 poll_users
[user
] = (struct fd
*)freelist
;
910 freelist
= &poll_users
[user
];
914 /* process pending timeouts and return the time until the next timeout, in milliseconds */
915 static int get_next_timeout(void)
917 int ret
= user_shared_data
? user_shared_data_timeout
: -1;
919 if (!list_empty( &abs_timeout_list
) || !list_empty( &rel_timeout_list
))
921 struct list expired_list
, *ptr
;
923 /* first remove all expired timers from the list */
925 list_init( &expired_list
);
926 while ((ptr
= list_head( &abs_timeout_list
)) != NULL
)
928 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
930 if (timeout
->when
<= current_time
)
932 list_remove( &timeout
->entry
);
933 list_add_tail( &expired_list
, &timeout
->entry
);
937 while ((ptr
= list_head( &rel_timeout_list
)) != NULL
)
939 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
941 if (-timeout
->when
<= monotonic_time
)
943 list_remove( &timeout
->entry
);
944 list_add_tail( &expired_list
, &timeout
->entry
);
949 /* now call the callback for all the removed timers */
951 while ((ptr
= list_head( &expired_list
)) != NULL
)
953 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
954 list_remove( &timeout
->entry
);
955 timeout
->callback( timeout
->private );
959 if ((ptr
= list_head( &abs_timeout_list
)) != NULL
)
961 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
962 timeout_t diff
= (timeout
->when
- current_time
+ 9999) / 10000;
963 if (diff
> INT_MAX
) diff
= INT_MAX
;
964 else if (diff
< 0) diff
= 0;
965 if (ret
== -1 || diff
< ret
) ret
= diff
;
968 if ((ptr
= list_head( &rel_timeout_list
)) != NULL
)
970 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
971 timeout_t diff
= (-timeout
->when
- monotonic_time
+ 9999) / 10000;
972 if (diff
> INT_MAX
) diff
= INT_MAX
;
973 else if (diff
< 0) diff
= 0;
974 if (ret
== -1 || diff
< ret
) ret
= diff
;
980 /* server main poll() loop */
986 server_start_time
= current_time
;
989 /* fall through to normal poll loop */
993 timeout
= get_next_timeout();
995 if (!active_users
) break; /* last user removed by a timeout */
997 ret
= poll( pollfd
, nb_users
, timeout
);
1002 for (i
= 0; i
< nb_users
; i
++)
1004 if (pollfd
[i
].revents
)
1006 fd_poll_event( poll_users
[i
], pollfd
[i
].revents
);
1015 /****************************************************************/
1016 /* device functions */
1018 static struct list device_hash
[DEVICE_HASH_SIZE
];
1020 static int is_device_removable( dev_t dev
, int unix_fd
)
1022 #if defined(linux) && defined(HAVE_FSTATFS)
1025 /* check for floppy disk */
1026 if (major(dev
) == FLOPPY_MAJOR
) return 1;
1028 if (fstatfs( unix_fd
, &stfs
) == -1) return 0;
1029 return (stfs
.f_type
== 0x9660 || /* iso9660 */
1030 stfs
.f_type
== 0x9fa1 || /* supermount */
1031 stfs
.f_type
== 0x15013346); /* udf */
1032 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__) || defined(__APPLE__)
1035 if (fstatfs( unix_fd
, &stfs
) == -1) return 0;
1036 return (!strcmp("cd9660", stfs
.f_fstypename
) || !strcmp("udf", stfs
.f_fstypename
));
1037 #elif defined(__NetBSD__)
1038 struct statvfs stfs
;
1040 if (fstatvfs( unix_fd
, &stfs
) == -1) return 0;
1041 return (!strcmp("cd9660", stfs
.f_fstypename
) || !strcmp("udf", stfs
.f_fstypename
));
1043 # include <sys/dkio.h>
1044 # include <sys/vtoc.h>
1045 struct dk_cinfo dkinf
;
1046 if (ioctl( unix_fd
, DKIOCINFO
, &dkinf
) == -1) return 0;
1047 return (dkinf
.dki_ctype
== DKC_CDROM
||
1048 dkinf
.dki_ctype
== DKC_NCRFLOPPY
||
1049 dkinf
.dki_ctype
== DKC_SMSFLOPPY
||
1050 dkinf
.dki_ctype
== DKC_INTEL82072
||
1051 dkinf
.dki_ctype
== DKC_INTEL82077
);
1057 /* retrieve the device object for a given fd, creating it if needed */
1058 static struct device
*get_device( dev_t dev
, int unix_fd
)
1060 struct device
*device
;
1061 unsigned int i
, hash
= dev
% DEVICE_HASH_SIZE
;
1063 if (device_hash
[hash
].next
)
1065 LIST_FOR_EACH_ENTRY( device
, &device_hash
[hash
], struct device
, entry
)
1066 if (device
->dev
== dev
) return (struct device
*)grab_object( device
);
1068 else list_init( &device_hash
[hash
] );
1070 /* not found, create it */
1072 if (unix_fd
== -1) return NULL
;
1073 if ((device
= alloc_object( &device_ops
)))
1076 device
->removable
= is_device_removable( dev
, unix_fd
);
1077 for (i
= 0; i
< INODE_HASH_SIZE
; i
++) list_init( &device
->inode_hash
[i
] );
1078 list_add_head( &device_hash
[hash
], &device
->entry
);
1083 static void device_dump( struct object
*obj
, int verbose
)
1085 struct device
*device
= (struct device
*)obj
;
1086 fprintf( stderr
, "Device dev=" );
1087 DUMP_LONG_LONG( device
->dev
);
1088 fprintf( stderr
, "\n" );
1091 static void device_destroy( struct object
*obj
)
1093 struct device
*device
= (struct device
*)obj
;
1096 for (i
= 0; i
< INODE_HASH_SIZE
; i
++)
1097 assert( list_empty(&device
->inode_hash
[i
]) );
1099 list_remove( &device
->entry
); /* remove it from the hash table */
1103 /****************************************************************/
1104 /* inode functions */
1106 /* close all pending file descriptors in the closed list */
1107 static void inode_close_pending( struct inode
*inode
, int keep_unlinks
)
1109 struct list
*ptr
= list_head( &inode
->closed
);
1113 struct closed_fd
*fd
= LIST_ENTRY( ptr
, struct closed_fd
, entry
);
1114 struct list
*next
= list_next( &inode
->closed
, ptr
);
1116 if (fd
->unix_fd
!= -1)
1118 close( fd
->unix_fd
);
1121 if (!keep_unlinks
|| !fd
->unlink
) /* get rid of it unless there's an unlink pending on that file */
1124 free( fd
->unix_name
);
1131 static void inode_dump( struct object
*obj
, int verbose
)
1133 struct inode
*inode
= (struct inode
*)obj
;
1134 fprintf( stderr
, "Inode device=%p ino=", inode
->device
);
1135 DUMP_LONG_LONG( inode
->ino
);
1136 fprintf( stderr
, "\n" );
1139 static void inode_destroy( struct object
*obj
)
1141 struct inode
*inode
= (struct inode
*)obj
;
1144 assert( list_empty(&inode
->open
) );
1145 assert( list_empty(&inode
->locks
) );
1147 list_remove( &inode
->entry
);
1149 while ((ptr
= list_head( &inode
->closed
)))
1151 struct closed_fd
*fd
= LIST_ENTRY( ptr
, struct closed_fd
, entry
);
1153 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1156 /* make sure it is still the same file */
1158 if (!stat( fd
->unix_name
, &st
) && st
.st_dev
== inode
->device
->dev
&& st
.st_ino
== inode
->ino
)
1160 if (S_ISDIR(st
.st_mode
)) rmdir( fd
->unix_name
);
1161 else unlink( fd
->unix_name
);
1164 free( fd
->unix_name
);
1167 release_object( inode
->device
);
1170 /* retrieve the inode object for a given fd, creating it if needed */
1171 static struct inode
*get_inode( dev_t dev
, ino_t ino
, int unix_fd
)
1173 struct device
*device
;
1174 struct inode
*inode
;
1175 unsigned int hash
= ino
% INODE_HASH_SIZE
;
1177 if (!(device
= get_device( dev
, unix_fd
))) return NULL
;
1179 LIST_FOR_EACH_ENTRY( inode
, &device
->inode_hash
[hash
], struct inode
, entry
)
1181 if (inode
->ino
== ino
)
1183 release_object( device
);
1184 return (struct inode
*)grab_object( inode
);
1188 /* not found, create it */
1189 if ((inode
= alloc_object( &inode_ops
)))
1191 inode
->device
= device
;
1193 list_init( &inode
->open
);
1194 list_init( &inode
->locks
);
1195 list_init( &inode
->closed
);
1196 list_add_head( &device
->inode_hash
[hash
], &inode
->entry
);
1198 else release_object( device
);
1203 /* add fd to the inode list of file descriptors to close */
1204 static void inode_add_closed_fd( struct inode
*inode
, struct closed_fd
*fd
)
1206 if (!list_empty( &inode
->locks
))
1208 list_add_head( &inode
->closed
, &fd
->entry
);
1210 else if (fd
->unlink
) /* close the fd but keep the structure around for unlink */
1212 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1214 list_add_head( &inode
->closed
, &fd
->entry
);
1216 else /* no locks on this inode and no unlink, get rid of the fd */
1218 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1219 free( fd
->unix_name
);
1225 /****************************************************************/
1226 /* file lock functions */
1228 static void file_lock_dump( struct object
*obj
, int verbose
)
1230 struct file_lock
*lock
= (struct file_lock
*)obj
;
1231 fprintf( stderr
, "Lock %s fd=%p proc=%p start=",
1232 lock
->shared
? "shared" : "excl", lock
->fd
, lock
->process
);
1233 DUMP_LONG_LONG( lock
->start
);
1234 fprintf( stderr
, " end=" );
1235 DUMP_LONG_LONG( lock
->end
);
1236 fprintf( stderr
, "\n" );
1239 static int file_lock_signaled( struct object
*obj
, struct wait_queue_entry
*entry
)
1241 struct file_lock
*lock
= (struct file_lock
*)obj
;
1242 /* lock is signaled if it has lost its owner */
1243 return !lock
->process
;
1246 /* set (or remove) a Unix lock if possible for the given range */
1247 static int set_unix_lock( struct fd
*fd
, file_pos_t start
, file_pos_t end
, int type
)
1251 if (!fd
->fs_locks
) return 1; /* no fs locks possible for this fd */
1254 if (start
== end
) return 1; /* can't set zero-byte lock */
1255 if (start
> max_unix_offset
) return 1; /* ignore it */
1257 fl
.l_whence
= SEEK_SET
;
1259 if (!end
|| end
> max_unix_offset
) fl
.l_len
= 0;
1260 else fl
.l_len
= end
- start
;
1261 if (fcntl( fd
->unix_fd
, F_SETLK
, &fl
) != -1) return 1;
1266 /* check whether locks work at all on this file system */
1267 if (fcntl( fd
->unix_fd
, F_GETLK
, &fl
) != -1)
1269 set_error( STATUS_FILE_LOCK_CONFLICT
);
1276 /* no locking on this fs, just ignore it */
1280 set_error( STATUS_FILE_LOCK_CONFLICT
);
1283 /* this can happen if we try to set a write lock on a read-only file */
1284 /* try to at least grab a read lock */
1285 if (fl
.l_type
== F_WRLCK
)
1290 set_error( STATUS_ACCESS_DENIED
);
1296 /* this can happen if off_t is 64-bit but the kernel only supports 32-bit */
1297 /* in that case we shrink the limit and retry */
1298 if (max_unix_offset
> INT_MAX
)
1300 max_unix_offset
= INT_MAX
;
1311 /* check if interval [start;end) overlaps the lock */
1312 static inline int lock_overlaps( struct file_lock
*lock
, file_pos_t start
, file_pos_t end
)
1314 if (lock
->end
&& start
>= lock
->end
) return 0;
1315 if (end
&& lock
->start
>= end
) return 0;
1319 /* remove Unix locks for all bytes in the specified area that are no longer locked */
1320 static void remove_unix_locks( struct fd
*fd
, file_pos_t start
, file_pos_t end
)
1328 } *first
, *cur
, *next
, *buffer
;
1333 if (!fd
->inode
) return;
1334 if (!fd
->fs_locks
) return;
1335 if (start
== end
|| start
> max_unix_offset
) return;
1336 if (!end
|| end
> max_unix_offset
) end
= max_unix_offset
+ 1;
1338 /* count the number of locks overlapping the specified area */
1340 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
1342 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
1343 if (lock
->start
== lock
->end
) continue;
1344 if (lock_overlaps( lock
, start
, end
)) count
++;
1347 if (!count
) /* no locks at all, we can unlock everything */
1349 set_unix_lock( fd
, start
, end
, F_UNLCK
);
1353 /* allocate space for the list of holes */
1354 /* max. number of holes is number of locks + 1 */
1356 if (!(buffer
= malloc( sizeof(*buffer
) * (count
+1) ))) return;
1360 first
->start
= start
;
1364 /* build a sorted list of unlocked holes in the specified area */
1366 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
1368 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
1369 if (lock
->start
== lock
->end
) continue;
1370 if (!lock_overlaps( lock
, start
, end
)) continue;
1372 /* go through all the holes touched by this lock */
1373 for (cur
= first
; cur
; cur
= cur
->next
)
1375 if (cur
->end
<= lock
->start
) continue; /* hole is before start of lock */
1376 if (lock
->end
&& cur
->start
>= lock
->end
) break; /* hole is after end of lock */
1378 /* now we know that lock is overlapping hole */
1380 if (cur
->start
>= lock
->start
) /* lock starts before hole, shrink from start */
1382 cur
->start
= lock
->end
;
1383 if (cur
->start
&& cur
->start
< cur
->end
) break; /* done with this lock */
1384 /* now hole is empty, remove it */
1385 if (cur
->next
) cur
->next
->prev
= cur
->prev
;
1386 if (cur
->prev
) cur
->prev
->next
= cur
->next
;
1387 else if (!(first
= cur
->next
)) goto done
; /* no more holes at all */
1389 else if (!lock
->end
|| cur
->end
<= lock
->end
) /* lock larger than hole, shrink from end */
1391 cur
->end
= lock
->start
;
1392 assert( cur
->start
< cur
->end
);
1394 else /* lock is in the middle of hole, split hole in two */
1397 next
->next
= cur
->next
;
1399 next
->start
= lock
->end
;
1400 next
->end
= cur
->end
;
1401 cur
->end
= lock
->start
;
1402 assert( next
->start
< next
->end
);
1403 assert( cur
->end
< next
->start
);
1405 break; /* done with this lock */
1410 /* clear Unix locks for all the holes */
1412 for (cur
= first
; cur
; cur
= cur
->next
)
1413 set_unix_lock( fd
, cur
->start
, cur
->end
, F_UNLCK
);
1419 /* create a new lock on a fd */
1420 static struct file_lock
*add_lock( struct fd
*fd
, int shared
, file_pos_t start
, file_pos_t end
)
1422 struct file_lock
*lock
;
1424 if (!(lock
= alloc_object( &file_lock_ops
))) return NULL
;
1425 lock
->shared
= shared
;
1426 lock
->start
= start
;
1429 lock
->process
= current
->process
;
1431 /* now try to set a Unix lock */
1432 if (!set_unix_lock( lock
->fd
, lock
->start
, lock
->end
, lock
->shared
? F_RDLCK
: F_WRLCK
))
1434 release_object( lock
);
1437 list_add_tail( &fd
->locks
, &lock
->fd_entry
);
1438 list_add_tail( &fd
->inode
->locks
, &lock
->inode_entry
);
1439 list_add_tail( &lock
->process
->locks
, &lock
->proc_entry
);
1443 /* remove an existing lock */
1444 static void remove_lock( struct file_lock
*lock
, int remove_unix
)
1446 struct inode
*inode
= lock
->fd
->inode
;
1448 list_remove( &lock
->fd_entry
);
1449 list_remove( &lock
->inode_entry
);
1450 list_remove( &lock
->proc_entry
);
1451 if (remove_unix
) remove_unix_locks( lock
->fd
, lock
->start
, lock
->end
);
1452 if (list_empty( &inode
->locks
)) inode_close_pending( inode
, 1 );
1453 lock
->process
= NULL
;
1454 wake_up( &lock
->obj
, 0 );
1455 release_object( lock
);
1458 /* remove all locks owned by a given process */
1459 void remove_process_locks( struct process
*process
)
1463 while ((ptr
= list_head( &process
->locks
)))
1465 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, proc_entry
);
1466 remove_lock( lock
, 1 ); /* this removes it from the list */
1470 /* remove all locks on a given fd */
1471 static void remove_fd_locks( struct fd
*fd
)
1473 file_pos_t start
= FILE_POS_T_MAX
, end
= 0;
1476 while ((ptr
= list_head( &fd
->locks
)))
1478 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, fd_entry
);
1479 if (lock
->start
< start
) start
= lock
->start
;
1480 if (!lock
->end
|| lock
->end
> end
) end
= lock
->end
- 1;
1481 remove_lock( lock
, 0 );
1483 if (start
< end
) remove_unix_locks( fd
, start
, end
+ 1 );
1486 /* add a lock on an fd */
1487 /* returns handle to wait on */
1488 obj_handle_t
lock_fd( struct fd
*fd
, file_pos_t start
, file_pos_t count
, int shared
, int wait
)
1491 file_pos_t end
= start
+ count
;
1493 if (!fd
->inode
) /* not a regular file */
1495 set_error( STATUS_INVALID_DEVICE_REQUEST
);
1499 /* don't allow wrapping locks */
1500 if (end
&& end
< start
)
1502 set_error( STATUS_INVALID_PARAMETER
);
1506 /* check if another lock on that file overlaps the area */
1507 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
1509 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
1510 if (!lock_overlaps( lock
, start
, end
)) continue;
1511 if (shared
&& (lock
->shared
|| lock
->fd
== fd
)) continue;
1515 set_error( STATUS_FILE_LOCK_CONFLICT
);
1518 set_error( STATUS_PENDING
);
1519 return alloc_handle( current
->process
, lock
, SYNCHRONIZE
, 0 );
1522 /* not found, add it */
1523 if (add_lock( fd
, shared
, start
, end
)) return 0;
1524 if (get_error() == STATUS_FILE_LOCK_CONFLICT
)
1526 /* Unix lock conflict -> tell client to wait and retry */
1527 if (wait
) set_error( STATUS_PENDING
);
1532 /* remove a lock on an fd */
1533 void unlock_fd( struct fd
*fd
, file_pos_t start
, file_pos_t count
)
1536 file_pos_t end
= start
+ count
;
1538 /* find an existing lock with the exact same parameters */
1539 LIST_FOR_EACH( ptr
, &fd
->locks
)
1541 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, fd_entry
);
1542 if ((lock
->start
== start
) && (lock
->end
== end
))
1544 remove_lock( lock
, 1 );
1548 set_error( STATUS_FILE_LOCK_CONFLICT
);
1552 /****************************************************************/
1553 /* file descriptor functions */
1555 static void fd_dump( struct object
*obj
, int verbose
)
1557 struct fd
*fd
= (struct fd
*)obj
;
1558 fprintf( stderr
, "Fd unix_fd=%d user=%p options=%08x", fd
->unix_fd
, fd
->user
, fd
->options
);
1559 if (fd
->inode
) fprintf( stderr
, " inode=%p unlink=%d", fd
->inode
, fd
->closed
->unlink
);
1560 fprintf( stderr
, "\n" );
1563 static void fd_destroy( struct object
*obj
)
1565 struct fd
*fd
= (struct fd
*)obj
;
1567 free_async_queue( &fd
->read_q
);
1568 free_async_queue( &fd
->write_q
);
1569 free_async_queue( &fd
->wait_q
);
1571 if (fd
->completion
) release_object( fd
->completion
);
1572 remove_fd_locks( fd
);
1573 list_remove( &fd
->inode_entry
);
1574 if (fd
->poll_index
!= -1) remove_poll_user( fd
, fd
->poll_index
);
1575 free( fd
->nt_name
);
1578 inode_add_closed_fd( fd
->inode
, fd
->closed
);
1579 release_object( fd
->inode
);
1581 else /* no inode, close it right away */
1583 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1584 free( fd
->unix_name
);
1588 /* check if the desired access is possible without violating */
1589 /* the sharing mode of other opens of the same file */
1590 static unsigned int check_sharing( struct fd
*fd
, unsigned int access
, unsigned int sharing
,
1591 unsigned int open_flags
, unsigned int options
)
1593 /* only a few access bits are meaningful wrt sharing */
1594 const unsigned int read_access
= FILE_READ_DATA
| FILE_EXECUTE
;
1595 const unsigned int write_access
= FILE_WRITE_DATA
| FILE_APPEND_DATA
;
1596 const unsigned int all_access
= read_access
| write_access
| DELETE
;
1598 unsigned int existing_sharing
= FILE_SHARE_READ
| FILE_SHARE_WRITE
| FILE_SHARE_DELETE
;
1599 unsigned int existing_access
= 0;
1602 fd
->access
= access
;
1603 fd
->sharing
= sharing
;
1605 LIST_FOR_EACH( ptr
, &fd
->inode
->open
)
1607 struct fd
*fd_ptr
= LIST_ENTRY( ptr
, struct fd
, inode_entry
);
1610 /* if access mode is 0, sharing mode is ignored */
1611 if (fd_ptr
->access
& all_access
) existing_sharing
&= fd_ptr
->sharing
;
1612 existing_access
|= fd_ptr
->access
;
1616 if (((access
& read_access
) && !(existing_sharing
& FILE_SHARE_READ
)) ||
1617 ((access
& write_access
) && !(existing_sharing
& FILE_SHARE_WRITE
)) ||
1618 ((access
& DELETE
) && !(existing_sharing
& FILE_SHARE_DELETE
)))
1619 return STATUS_SHARING_VIOLATION
;
1620 if (((existing_access
& FILE_MAPPING_WRITE
) && !(sharing
& FILE_SHARE_WRITE
)) ||
1621 ((existing_access
& FILE_MAPPING_IMAGE
) && (access
& FILE_WRITE_DATA
)))
1622 return STATUS_SHARING_VIOLATION
;
1623 if ((existing_access
& FILE_MAPPING_IMAGE
) && (options
& FILE_DELETE_ON_CLOSE
))
1624 return STATUS_CANNOT_DELETE
;
1625 if ((existing_access
& FILE_MAPPING_ACCESS
) && (open_flags
& O_TRUNC
))
1626 return STATUS_USER_MAPPED_FILE
;
1627 if (!(access
& all_access
))
1628 return 0; /* if access mode is 0, sharing mode is ignored (except for mappings) */
1629 if (((existing_access
& read_access
) && !(sharing
& FILE_SHARE_READ
)) ||
1630 ((existing_access
& write_access
) && !(sharing
& FILE_SHARE_WRITE
)) ||
1631 ((existing_access
& DELETE
) && !(sharing
& FILE_SHARE_DELETE
)))
1632 return STATUS_SHARING_VIOLATION
;
1636 /* set the events that select waits for on this fd */
1637 void set_fd_events( struct fd
*fd
, int events
)
1639 int user
= fd
->poll_index
;
1640 assert( poll_users
[user
] == fd
);
1642 set_fd_epoll_events( fd
, user
, events
);
1644 if (events
== -1) /* stop waiting on this fd completely */
1646 pollfd
[user
].fd
= -1;
1647 pollfd
[user
].events
= POLLERR
;
1648 pollfd
[user
].revents
= 0;
1650 else if (pollfd
[user
].fd
!= -1 || !pollfd
[user
].events
)
1652 pollfd
[user
].fd
= fd
->unix_fd
;
1653 pollfd
[user
].events
= events
;
1657 /* prepare an fd for unmounting its corresponding device */
1658 static inline void unmount_fd( struct fd
*fd
)
1660 assert( fd
->inode
);
1662 async_wake_up( &fd
->read_q
, STATUS_VOLUME_DISMOUNTED
);
1663 async_wake_up( &fd
->write_q
, STATUS_VOLUME_DISMOUNTED
);
1665 if (fd
->poll_index
!= -1) set_fd_events( fd
, -1 );
1667 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1670 fd
->no_fd_status
= STATUS_VOLUME_DISMOUNTED
;
1671 fd
->closed
->unix_fd
= -1;
1672 fd
->closed
->unlink
= 0;
1674 /* stop using Unix locks on this fd (existing locks have been removed by close) */
1678 /* allocate an fd object, without setting the unix fd yet */
1679 static struct fd
*alloc_fd_object(void)
1681 struct fd
*fd
= alloc_object( &fd_ops
);
1683 if (!fd
) return NULL
;
1693 fd
->unix_name
= NULL
;
1699 fd
->poll_index
= -1;
1700 fd
->completion
= NULL
;
1702 init_async_queue( &fd
->read_q
);
1703 init_async_queue( &fd
->write_q
);
1704 init_async_queue( &fd
->wait_q
);
1705 list_init( &fd
->inode_entry
);
1706 list_init( &fd
->locks
);
1708 if ((fd
->poll_index
= add_poll_user( fd
)) == -1)
1710 release_object( fd
);
1716 /* allocate a pseudo fd object, for objects that need to behave like files but don't have a unix fd */
1717 struct fd
*alloc_pseudo_fd( const struct fd_ops
*fd_user_ops
, struct object
*user
, unsigned int options
)
1719 struct fd
*fd
= alloc_object( &fd_ops
);
1721 if (!fd
) return NULL
;
1723 fd
->fd_ops
= fd_user_ops
;
1728 fd
->options
= options
;
1730 fd
->unix_name
= NULL
;
1737 fd
->poll_index
= -1;
1738 fd
->completion
= NULL
;
1740 fd
->no_fd_status
= STATUS_BAD_DEVICE_TYPE
;
1741 init_async_queue( &fd
->read_q
);
1742 init_async_queue( &fd
->write_q
);
1743 init_async_queue( &fd
->wait_q
);
1744 list_init( &fd
->inode_entry
);
1745 list_init( &fd
->locks
);
1749 /* duplicate an fd object for a different user */
1750 struct fd
*dup_fd_object( struct fd
*orig
, unsigned int access
, unsigned int sharing
, unsigned int options
)
1753 struct fd
*fd
= alloc_fd_object();
1755 if (!fd
) return NULL
;
1757 fd
->options
= options
;
1758 fd
->cacheable
= orig
->cacheable
;
1760 if (orig
->unix_name
)
1762 if (!(fd
->unix_name
= mem_alloc( strlen(orig
->unix_name
) + 1 ))) goto failed
;
1763 strcpy( fd
->unix_name
, orig
->unix_name
);
1765 if (orig
->nt_namelen
)
1767 if (!(fd
->nt_name
= memdup( orig
->nt_name
, orig
->nt_namelen
))) goto failed
;
1768 fd
->nt_namelen
= orig
->nt_namelen
;
1773 struct closed_fd
*closed
= mem_alloc( sizeof(*closed
) );
1774 if (!closed
) goto failed
;
1775 if ((fd
->unix_fd
= dup( orig
->unix_fd
)) == -1)
1781 closed
->unix_fd
= fd
->unix_fd
;
1783 closed
->unix_name
= fd
->unix_name
;
1784 fd
->closed
= closed
;
1785 fd
->inode
= (struct inode
*)grab_object( orig
->inode
);
1786 list_add_head( &fd
->inode
->open
, &fd
->inode_entry
);
1787 if ((err
= check_sharing( fd
, access
, sharing
, 0, options
)))
1793 else if ((fd
->unix_fd
= dup( orig
->unix_fd
)) == -1)
1801 release_object( fd
);
1805 /* find an existing fd object that can be reused for a mapping */
1806 struct fd
*get_fd_object_for_mapping( struct fd
*fd
, unsigned int access
, unsigned int sharing
)
1810 if (!fd
->inode
) return NULL
;
1812 LIST_FOR_EACH_ENTRY( fd_ptr
, &fd
->inode
->open
, struct fd
, inode_entry
)
1813 if (fd_ptr
->access
== access
&& fd_ptr
->sharing
== sharing
)
1814 return (struct fd
*)grab_object( fd_ptr
);
1819 /* sets the user of an fd that previously had no user */
1820 void set_fd_user( struct fd
*fd
, const struct fd_ops
*user_ops
, struct object
*user
)
1822 assert( fd
->fd_ops
== NULL
);
1823 fd
->fd_ops
= user_ops
;
1827 char *dup_fd_name( struct fd
*root
, const char *name
)
1831 if (!root
) return strdup( name
);
1832 if (!root
->unix_name
) return NULL
;
1835 if (name
[0] == '.' && (!name
[1] || name
[1] == '/')) name
++;
1837 if ((ret
= malloc( strlen(root
->unix_name
) + strlen(name
) + 2 )))
1839 strcpy( ret
, root
->unix_name
);
1840 if (name
[0] && name
[0] != '/') strcat( ret
, "/" );
1841 strcat( ret
, name
);
1846 static WCHAR
*dup_nt_name( struct fd
*root
, struct unicode_str name
, data_size_t
*len
)
1854 if (!name
.len
) return NULL
;
1855 return memdup( name
.str
, name
.len
);
1857 if (!root
->nt_namelen
) return NULL
;
1858 retlen
= root
->nt_namelen
;
1861 if (name
.len
&& name
.str
[0] == '.' && (name
.len
== sizeof(WCHAR
) || name
.str
[1] == '\\'))
1864 name
.len
-= sizeof(WCHAR
);
1866 if ((ret
= malloc( retlen
+ name
.len
+ 1 )))
1868 memcpy( ret
, root
->nt_name
, root
->nt_namelen
);
1869 if (name
.len
&& name
.str
[0] != '\\' &&
1870 root
->nt_namelen
&& root
->nt_name
[root
->nt_namelen
/ sizeof(WCHAR
) - 1] != '\\')
1872 ret
[retlen
/ sizeof(WCHAR
)] = '\\';
1873 retlen
+= sizeof(WCHAR
);
1875 memcpy( ret
+ retlen
/ sizeof(WCHAR
), name
.str
, name
.len
);
1876 *len
= retlen
+ name
.len
;
1881 void get_nt_name( struct fd
*fd
, struct unicode_str
*name
)
1883 name
->str
= fd
->nt_name
;
1884 name
->len
= fd
->nt_namelen
;
1887 /* open() wrapper that returns a struct fd with no fd user set */
1888 struct fd
*open_fd( struct fd
*root
, const char *name
, struct unicode_str nt_name
,
1889 int flags
, mode_t
*mode
, unsigned int access
,
1890 unsigned int sharing
, unsigned int options
)
1893 struct closed_fd
*closed_fd
;
1899 if (((options
& FILE_DELETE_ON_CLOSE
) && !(access
& DELETE
)) ||
1900 ((options
& FILE_DIRECTORY_FILE
) && (flags
& O_TRUNC
)))
1902 set_error( STATUS_INVALID_PARAMETER
);
1906 if (!(fd
= alloc_fd_object())) return NULL
;
1908 fd
->options
= options
;
1909 if (!(closed_fd
= mem_alloc( sizeof(*closed_fd
) )))
1911 release_object( fd
);
1917 if ((root_fd
= get_unix_fd( root
)) == -1) goto error
;
1918 if (fchdir( root_fd
) == -1)
1926 /* create the directory if needed */
1927 if ((options
& FILE_DIRECTORY_FILE
) && (flags
& O_CREAT
))
1929 if (mkdir( name
, *mode
) == -1)
1931 if (errno
!= EEXIST
|| (flags
& O_EXCL
))
1937 flags
&= ~(O_CREAT
| O_EXCL
| O_TRUNC
);
1940 if ((access
& FILE_UNIX_WRITE_ACCESS
) && !(options
& FILE_DIRECTORY_FILE
))
1942 if (access
& FILE_UNIX_READ_ACCESS
) rw_mode
= O_RDWR
;
1943 else rw_mode
= O_WRONLY
;
1945 else rw_mode
= O_RDONLY
;
1947 if ((fd
->unix_fd
= open( name
, rw_mode
| (flags
& ~O_TRUNC
), *mode
)) == -1)
1949 /* if we tried to open a directory for write access, retry read-only */
1950 if (errno
== EISDIR
)
1952 if ((access
& FILE_UNIX_WRITE_ACCESS
) || (flags
& O_CREAT
))
1953 fd
->unix_fd
= open( name
, O_RDONLY
| (flags
& ~(O_TRUNC
| O_CREAT
| O_EXCL
)), *mode
);
1956 if (fd
->unix_fd
== -1)
1963 fd
->nt_name
= dup_nt_name( root
, nt_name
, &fd
->nt_namelen
);
1964 fd
->unix_name
= NULL
;
1965 if ((path
= dup_fd_name( root
, name
)))
1967 fd
->unix_name
= realpath( path
, NULL
);
1971 closed_fd
->unix_fd
= fd
->unix_fd
;
1972 closed_fd
->unlink
= 0;
1973 closed_fd
->unix_name
= fd
->unix_name
;
1974 fstat( fd
->unix_fd
, &st
);
1977 /* only bother with an inode for normal files and directories */
1978 if (S_ISREG(st
.st_mode
) || S_ISDIR(st
.st_mode
))
1981 struct inode
*inode
= get_inode( st
.st_dev
, st
.st_ino
, fd
->unix_fd
);
1985 /* we can close the fd because there are no others open on the same file,
1986 * otherwise we wouldn't have failed to allocate a new inode
1991 fd
->closed
= closed_fd
;
1992 fd
->cacheable
= !inode
->device
->removable
;
1993 list_add_head( &inode
->open
, &fd
->inode_entry
);
1996 /* check directory options */
1997 if ((options
& FILE_DIRECTORY_FILE
) && !S_ISDIR(st
.st_mode
))
1999 set_error( STATUS_NOT_A_DIRECTORY
);
2002 if ((options
& FILE_NON_DIRECTORY_FILE
) && S_ISDIR(st
.st_mode
))
2004 set_error( STATUS_FILE_IS_A_DIRECTORY
);
2007 if ((err
= check_sharing( fd
, access
, sharing
, flags
, options
)))
2013 /* can't unlink files if we don't have permission to access */
2014 if ((options
& FILE_DELETE_ON_CLOSE
) && !(flags
& O_CREAT
) &&
2015 !(st
.st_mode
& (S_IWUSR
| S_IWGRP
| S_IWOTH
)))
2017 set_error( STATUS_CANNOT_DELETE
);
2021 fd
->closed
->unlink
= (options
& FILE_DELETE_ON_CLOSE
) ? -1 : 0;
2022 if (flags
& O_TRUNC
)
2024 if (S_ISDIR(st
.st_mode
))
2026 set_error( STATUS_OBJECT_NAME_COLLISION
);
2029 ftruncate( fd
->unix_fd
, 0 );
2032 else /* special file */
2034 if (options
& FILE_DELETE_ON_CLOSE
) /* we can't unlink special files */
2036 set_error( STATUS_INVALID_PARAMETER
);
2042 if (root_fd
!= -1) fchdir( server_dir_fd
); /* go back to the server dir */
2046 release_object( fd
);
2048 if (root_fd
!= -1) fchdir( server_dir_fd
); /* go back to the server dir */
2052 /* create an fd for an anonymous file */
2053 /* if the function fails the unix fd is closed */
2054 struct fd
*create_anonymous_fd( const struct fd_ops
*fd_user_ops
, int unix_fd
, struct object
*user
,
2055 unsigned int options
)
2057 struct fd
*fd
= alloc_fd_object();
2061 set_fd_user( fd
, fd_user_ops
, user
);
2062 fd
->unix_fd
= unix_fd
;
2063 fd
->options
= options
;
2070 /* retrieve the object that is using an fd */
2071 void *get_fd_user( struct fd
*fd
)
2076 /* retrieve the opening options for the fd */
2077 unsigned int get_fd_options( struct fd
*fd
)
2082 /* check if fd is in overlapped mode */
2083 int is_fd_overlapped( struct fd
*fd
)
2085 return !(fd
->options
& (FILE_SYNCHRONOUS_IO_ALERT
| FILE_SYNCHRONOUS_IO_NONALERT
));
2088 /* retrieve the unix fd for an object */
2089 int get_unix_fd( struct fd
*fd
)
2091 if (fd
->unix_fd
== -1) set_error( fd
->no_fd_status
);
2095 /* check if two file descriptors point to the same file */
2096 int is_same_file_fd( struct fd
*fd1
, struct fd
*fd2
)
2098 return fd1
->inode
== fd2
->inode
;
2101 /* allow the fd to be cached (can't be reset once set) */
2102 void allow_fd_caching( struct fd
*fd
)
2107 /* check if fd is on a removable device */
2108 int is_fd_removable( struct fd
*fd
)
2110 return (fd
->inode
&& fd
->inode
->device
->removable
);
2113 /* set or clear the fd signaled state */
2114 void set_fd_signaled( struct fd
*fd
, int signaled
)
2116 if (fd
->comp_flags
& FILE_SKIP_SET_EVENT_ON_HANDLE
) return;
2117 fd
->signaled
= signaled
;
2118 if (signaled
) wake_up( fd
->user
, 0 );
2121 /* check if events are pending and if yes return which one(s) */
2122 int check_fd_events( struct fd
*fd
, int events
)
2126 if (fd
->unix_fd
== -1) return POLLERR
;
2127 if (fd
->inode
) return events
; /* regular files are always signaled */
2129 pfd
.fd
= fd
->unix_fd
;
2130 pfd
.events
= events
;
2131 if (poll( &pfd
, 1, 0 ) <= 0) return 0;
2135 /* default signaled() routine for objects that poll() on an fd */
2136 int default_fd_signaled( struct object
*obj
, struct wait_queue_entry
*entry
)
2138 struct fd
*fd
= get_obj_fd( obj
);
2139 int ret
= fd
->signaled
;
2140 release_object( fd
);
2144 int default_fd_get_poll_events( struct fd
*fd
)
2148 if (async_waiting( &fd
->read_q
)) events
|= POLLIN
;
2149 if (async_waiting( &fd
->write_q
)) events
|= POLLOUT
;
2153 /* default handler for poll() events */
2154 void default_poll_event( struct fd
*fd
, int event
)
2156 if (event
& (POLLIN
| POLLERR
| POLLHUP
)) async_wake_up( &fd
->read_q
, STATUS_ALERTED
);
2157 if (event
& (POLLOUT
| POLLERR
| POLLHUP
)) async_wake_up( &fd
->write_q
, STATUS_ALERTED
);
2159 /* if an error occurred, stop polling this fd to avoid busy-looping */
2160 if (event
& (POLLERR
| POLLHUP
)) set_fd_events( fd
, -1 );
2161 else if (!fd
->inode
) set_fd_events( fd
, fd
->fd_ops
->get_poll_events( fd
) );
2164 void fd_queue_async( struct fd
*fd
, struct async
*async
, int type
)
2166 struct async_queue
*queue
;
2170 case ASYNC_TYPE_READ
:
2171 queue
= &fd
->read_q
;
2173 case ASYNC_TYPE_WRITE
:
2174 queue
= &fd
->write_q
;
2176 case ASYNC_TYPE_WAIT
:
2177 queue
= &fd
->wait_q
;
2184 queue_async( queue
, async
);
2186 if (type
!= ASYNC_TYPE_WAIT
)
2189 set_fd_events( fd
, fd
->fd_ops
->get_poll_events( fd
) );
2190 else /* regular files are always ready for read and write */
2191 async_wake_up( queue
, STATUS_ALERTED
);
2195 void fd_async_wake_up( struct fd
*fd
, int type
, unsigned int status
)
2199 case ASYNC_TYPE_READ
:
2200 async_wake_up( &fd
->read_q
, status
);
2202 case ASYNC_TYPE_WRITE
:
2203 async_wake_up( &fd
->write_q
, status
);
2205 case ASYNC_TYPE_WAIT
:
2206 async_wake_up( &fd
->wait_q
, status
);
2213 void fd_reselect_async( struct fd
*fd
, struct async_queue
*queue
)
2215 fd
->fd_ops
->reselect_async( fd
, queue
);
2218 void no_fd_queue_async( struct fd
*fd
, struct async
*async
, int type
, int count
)
2220 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2223 void default_fd_queue_async( struct fd
*fd
, struct async
*async
, int type
, int count
)
2225 fd_queue_async( fd
, async
, type
);
2226 set_error( STATUS_PENDING
);
2229 /* default reselect_async() fd routine */
2230 void default_fd_reselect_async( struct fd
*fd
, struct async_queue
*queue
)
2232 if (queue
== &fd
->read_q
|| queue
== &fd
->write_q
)
2234 int poll_events
= fd
->fd_ops
->get_poll_events( fd
);
2235 int events
= check_fd_events( fd
, poll_events
);
2236 if (events
) fd
->fd_ops
->poll_event( fd
, events
);
2237 else set_fd_events( fd
, poll_events
);
2241 static inline int is_valid_mounted_device( struct stat
*st
)
2243 #if defined(linux) || defined(__sun__)
2244 return S_ISBLK( st
->st_mode
);
2246 /* disks are char devices on *BSD */
2247 return S_ISCHR( st
->st_mode
);
2251 /* close all Unix file descriptors on a device to allow unmounting it */
2252 static void unmount_device( struct fd
*device_fd
)
2256 struct device
*device
;
2257 struct inode
*inode
;
2259 int unix_fd
= get_unix_fd( device_fd
);
2261 if (unix_fd
== -1) return;
2263 if (fstat( unix_fd
, &st
) == -1 || !is_valid_mounted_device( &st
))
2265 set_error( STATUS_INVALID_PARAMETER
);
2269 if (!(device
= get_device( st
.st_rdev
, -1 ))) return;
2271 for (i
= 0; i
< INODE_HASH_SIZE
; i
++)
2273 LIST_FOR_EACH_ENTRY( inode
, &device
->inode_hash
[i
], struct inode
, entry
)
2275 LIST_FOR_EACH_ENTRY( fd
, &inode
->open
, struct fd
, inode_entry
)
2279 inode_close_pending( inode
, 0 );
2282 /* remove it from the hash table */
2283 list_remove( &device
->entry
);
2284 list_init( &device
->entry
);
2285 release_object( device
);
2288 /* default read() routine */
2289 int no_fd_read( struct fd
*fd
, struct async
*async
, file_pos_t pos
)
2291 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2295 /* default write() routine */
2296 int no_fd_write( struct fd
*fd
, struct async
*async
, file_pos_t pos
)
2298 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2302 /* default flush() routine */
2303 int no_fd_flush( struct fd
*fd
, struct async
*async
)
2305 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2309 /* default get_file_info() routine */
2310 void no_fd_get_file_info( struct fd
*fd
, obj_handle_t handle
, unsigned int info_class
)
2312 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2315 /* default get_file_info() routine */
2316 void default_fd_get_file_info( struct fd
*fd
, obj_handle_t handle
, unsigned int info_class
)
2320 case FileAccessInformation
:
2322 FILE_ACCESS_INFORMATION info
;
2323 if (get_reply_max_size() < sizeof(info
))
2325 set_error( STATUS_INFO_LENGTH_MISMATCH
);
2328 info
.AccessFlags
= get_handle_access( current
->process
, handle
);
2329 set_reply_data( &info
, sizeof(info
) );
2332 case FileModeInformation
:
2334 FILE_MODE_INFORMATION info
;
2335 if (get_reply_max_size() < sizeof(info
))
2337 set_error( STATUS_INFO_LENGTH_MISMATCH
);
2340 info
.Mode
= fd
->options
& ( FILE_WRITE_THROUGH
2341 | FILE_SEQUENTIAL_ONLY
2342 | FILE_NO_INTERMEDIATE_BUFFERING
2343 | FILE_SYNCHRONOUS_IO_ALERT
2344 | FILE_SYNCHRONOUS_IO_NONALERT
);
2345 set_reply_data( &info
, sizeof(info
) );
2348 case FileIoCompletionNotificationInformation
:
2350 FILE_IO_COMPLETION_NOTIFICATION_INFORMATION info
;
2351 if (get_reply_max_size() < sizeof(info
))
2353 set_error( STATUS_INFO_LENGTH_MISMATCH
);
2356 info
.Flags
= fd
->comp_flags
;
2357 set_reply_data( &info
, sizeof(info
) );
2361 set_error( STATUS_NOT_IMPLEMENTED
);
2365 /* default get_volume_info() routine */
2366 int no_fd_get_volume_info( struct fd
*fd
, struct async
*async
, unsigned int info_class
)
2368 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2372 /* default ioctl() routine */
2373 int no_fd_ioctl( struct fd
*fd
, ioctl_code_t code
, struct async
*async
)
2375 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2379 /* default ioctl() routine */
2380 int default_fd_ioctl( struct fd
*fd
, ioctl_code_t code
, struct async
*async
)
2384 case FSCTL_DISMOUNT_VOLUME
:
2385 unmount_device( fd
);
2388 set_error( STATUS_NOT_SUPPORTED
);
2393 /* same as get_handle_obj but retrieve the struct fd associated to the object */
2394 static struct fd
*get_handle_fd_obj( struct process
*process
, obj_handle_t handle
,
2395 unsigned int access
)
2397 struct fd
*fd
= NULL
;
2400 if ((obj
= get_handle_obj( process
, handle
, access
, NULL
)))
2402 fd
= get_obj_fd( obj
);
2403 release_object( obj
);
2408 static int is_dir_empty( int fd
)
2414 if ((fd
= dup( fd
)) == -1)
2417 if (!(dir
= fdopendir( fd
)))
2424 while (empty
&& (de
= readdir( dir
)))
2426 if (!strcmp( de
->d_name
, "." ) || !strcmp( de
->d_name
, ".." )) continue;
2433 /* set disposition for the fd */
2434 static void set_fd_disposition( struct fd
*fd
, int unlink
)
2440 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2444 if (fd
->unix_fd
== -1)
2446 set_error( fd
->no_fd_status
);
2452 if (fstat( fd
->unix_fd
, &st
) == -1)
2457 if (S_ISREG( st
.st_mode
)) /* can't unlink files we don't have permission to write */
2459 if (!(st
.st_mode
& (S_IWUSR
| S_IWGRP
| S_IWOTH
)))
2461 set_error( STATUS_CANNOT_DELETE
);
2465 else if (S_ISDIR( st
.st_mode
)) /* can't remove non-empty directories */
2467 switch (is_dir_empty( fd
->unix_fd
))
2473 set_error( STATUS_DIRECTORY_NOT_EMPTY
);
2477 else /* can't unlink special files */
2479 set_error( STATUS_INVALID_PARAMETER
);
2484 fd
->closed
->unlink
= unlink
? 1 : 0;
2485 if (fd
->options
& FILE_DELETE_ON_CLOSE
)
2486 fd
->closed
->unlink
= -1;
2489 /* set new name for the fd */
2490 static void set_fd_name( struct fd
*fd
, struct fd
*root
, const char *nameptr
, data_size_t len
,
2491 struct unicode_str nt_name
, int create_link
, int replace
)
2493 struct inode
*inode
;
2494 struct stat st
, st2
;
2497 if (!fd
->inode
|| !fd
->unix_name
)
2499 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2502 if (fd
->unix_fd
== -1)
2504 set_error( fd
->no_fd_status
);
2508 if (!len
|| ((nameptr
[0] == '/') ^ !root
))
2510 set_error( STATUS_OBJECT_PATH_SYNTAX_BAD
);
2513 if (!(name
= mem_alloc( len
+ 1 ))) return;
2514 memcpy( name
, nameptr
, len
);
2519 char *combined_name
= dup_fd_name( root
, name
);
2522 set_error( STATUS_NO_MEMORY
);
2526 name
= combined_name
;
2529 /* when creating a hard link, source cannot be a dir */
2530 if (create_link
&& !fstat( fd
->unix_fd
, &st
) && S_ISDIR( st
.st_mode
))
2532 set_error( STATUS_FILE_IS_A_DIRECTORY
);
2536 if (!stat( name
, &st
))
2538 if (!fstat( fd
->unix_fd
, &st2
) && st
.st_ino
== st2
.st_ino
&& st
.st_dev
== st2
.st_dev
)
2540 if (create_link
&& !replace
) set_error( STATUS_OBJECT_NAME_COLLISION
);
2547 set_error( STATUS_OBJECT_NAME_COLLISION
);
2551 /* can't replace directories or special files */
2552 if (!S_ISREG( st
.st_mode
))
2554 set_error( STATUS_ACCESS_DENIED
);
2558 /* can't replace an opened file */
2559 if ((inode
= get_inode( st
.st_dev
, st
.st_ino
, -1 )))
2561 int is_empty
= list_empty( &inode
->open
);
2562 release_object( inode
);
2565 set_error( STATUS_ACCESS_DENIED
);
2570 /* link() expects that the target doesn't exist */
2571 /* rename() cannot replace files with directories */
2572 if (create_link
|| S_ISDIR( st2
.st_mode
))
2584 if (link( fd
->unix_name
, name
))
2590 if (rename( fd
->unix_name
, name
))
2596 if (is_file_executable( fd
->unix_name
) != is_file_executable( name
) && !fstat( fd
->unix_fd
, &st
))
2598 if (is_file_executable( name
))
2599 /* set executable bit where read bit is set */
2600 st
.st_mode
|= (st
.st_mode
& 0444) >> 2;
2602 st
.st_mode
&= ~0111;
2603 fchmod( fd
->unix_fd
, st
.st_mode
);
2606 free( fd
->nt_name
);
2607 fd
->nt_name
= dup_nt_name( root
, nt_name
, &fd
->nt_namelen
);
2608 free( fd
->unix_name
);
2609 fd
->closed
->unix_name
= fd
->unix_name
= realpath( name
, NULL
);
2612 set_error( STATUS_NO_MEMORY
);
2619 struct completion
*fd_get_completion( struct fd
*fd
, apc_param_t
*p_key
)
2621 *p_key
= fd
->comp_key
;
2622 return fd
->completion
? (struct completion
*)grab_object( fd
->completion
) : NULL
;
2625 void fd_copy_completion( struct fd
*src
, struct fd
*dst
)
2627 assert( !dst
->completion
);
2628 dst
->completion
= fd_get_completion( src
, &dst
->comp_key
);
2629 dst
->comp_flags
= src
->comp_flags
;
2632 /* flush a file buffers */
2635 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->async
.handle
, 0 );
2636 struct async
*async
;
2640 if ((async
= create_request_async( fd
, fd
->comp_flags
, &req
->async
)))
2642 reply
->event
= async_handoff( async
, fd
->fd_ops
->flush( fd
, async
), NULL
, 1 );
2643 release_object( async
);
2645 release_object( fd
);
2648 /* query file info */
2649 DECL_HANDLER(get_file_info
)
2651 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
2655 fd
->fd_ops
->get_file_info( fd
, req
->handle
, req
->info_class
);
2656 release_object( fd
);
2660 /* query volume info */
2661 DECL_HANDLER(get_volume_info
)
2663 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
2664 struct async
*async
;
2668 if ((async
= create_request_async( fd
, fd
->comp_flags
, &req
->async
)))
2670 reply
->wait
= async_handoff( async
, fd
->fd_ops
->get_volume_info( fd
, async
, req
->info_class
), NULL
, 1 );
2671 release_object( async
);
2673 release_object( fd
);
2676 /* open a file object */
2677 DECL_HANDLER(open_file_object
)
2679 struct unicode_str name
= get_req_unicode_str();
2680 struct object
*obj
, *result
, *root
= NULL
;
2682 if (req
->rootdir
&& !(root
= get_handle_obj( current
->process
, req
->rootdir
, 0, NULL
))) return;
2684 obj
= open_named_object( root
, NULL
, &name
, req
->attributes
);
2685 if (root
) release_object( root
);
2688 if ((result
= obj
->ops
->open_file( obj
, req
->access
, req
->sharing
, req
->options
)))
2690 reply
->handle
= alloc_handle( current
->process
, result
, req
->access
, req
->attributes
);
2691 release_object( result
);
2693 release_object( obj
);
2696 /* get the Unix name from a file handle */
2697 DECL_HANDLER(get_handle_unix_name
)
2701 if ((fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 )))
2705 data_size_t name_len
= strlen( fd
->unix_name
);
2706 reply
->name_len
= name_len
;
2707 if (name_len
<= get_reply_max_size()) set_reply_data( fd
->unix_name
, name_len
);
2708 else set_error( STATUS_BUFFER_OVERFLOW
);
2710 else set_error( STATUS_OBJECT_TYPE_MISMATCH
);
2711 release_object( fd
);
2715 /* get a Unix fd to access a file */
2716 DECL_HANDLER(get_handle_fd
)
2720 if ((fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 )))
2722 int unix_fd
= get_unix_fd( fd
);
2723 reply
->cacheable
= fd
->cacheable
;
2726 reply
->type
= fd
->fd_ops
->get_fd_type( fd
);
2727 reply
->options
= fd
->options
;
2728 reply
->access
= get_handle_access( current
->process
, req
->handle
);
2729 send_client_fd( current
->process
, unix_fd
, req
->handle
);
2731 release_object( fd
);
2735 /* perform a read on a file object */
2738 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->async
.handle
, FILE_READ_DATA
);
2739 struct async
*async
;
2743 if ((async
= create_request_async( fd
, fd
->comp_flags
, &req
->async
)))
2745 reply
->wait
= async_handoff( async
, fd
->fd_ops
->read( fd
, async
, req
->pos
), NULL
, 0 );
2746 reply
->options
= fd
->options
;
2747 release_object( async
);
2749 release_object( fd
);
2752 /* perform a write on a file object */
2755 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->async
.handle
, FILE_WRITE_DATA
);
2756 struct async
*async
;
2760 if ((async
= create_request_async( fd
, fd
->comp_flags
, &req
->async
)))
2762 reply
->wait
= async_handoff( async
, fd
->fd_ops
->write( fd
, async
, req
->pos
), &reply
->size
, 0 );
2763 reply
->options
= fd
->options
;
2764 release_object( async
);
2766 release_object( fd
);
2769 /* perform an ioctl on a file */
2772 unsigned int access
= (req
->code
>> 14) & (FILE_READ_DATA
|FILE_WRITE_DATA
);
2773 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->async
.handle
, access
);
2774 struct async
*async
;
2778 if ((async
= create_request_async( fd
, fd
->comp_flags
, &req
->async
)))
2780 reply
->wait
= async_handoff( async
, fd
->fd_ops
->ioctl( fd
, req
->code
, async
), NULL
, 0 );
2781 reply
->options
= fd
->options
;
2782 release_object( async
);
2784 release_object( fd
);
2787 /* create / reschedule an async I/O */
2788 DECL_HANDLER(register_async
)
2790 unsigned int access
;
2791 struct async
*async
;
2796 case ASYNC_TYPE_READ
:
2797 access
= FILE_READ_DATA
;
2799 case ASYNC_TYPE_WRITE
:
2800 access
= FILE_WRITE_DATA
;
2803 set_error( STATUS_INVALID_PARAMETER
);
2807 if ((fd
= get_handle_fd_obj( current
->process
, req
->async
.handle
, access
)))
2809 if (get_unix_fd( fd
) != -1 && (async
= create_async( fd
, current
, &req
->async
, NULL
)))
2811 fd
->fd_ops
->queue_async( fd
, async
, req
->type
, req
->count
);
2812 release_object( async
);
2814 release_object( fd
);
2818 /* attach completion object to a fd */
2819 DECL_HANDLER(set_completion_info
)
2821 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
2825 if (is_fd_overlapped( fd
) && !fd
->completion
)
2827 fd
->completion
= get_completion_obj( current
->process
, req
->chandle
, IO_COMPLETION_MODIFY_STATE
);
2828 fd
->comp_key
= req
->ckey
;
2830 else set_error( STATUS_INVALID_PARAMETER
);
2831 release_object( fd
);
2835 /* push new completion msg into a completion queue attached to the fd */
2836 DECL_HANDLER(add_fd_completion
)
2838 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
2841 if (fd
->completion
&& (req
->async
|| !(fd
->comp_flags
& FILE_SKIP_COMPLETION_PORT_ON_SUCCESS
)))
2842 add_completion( fd
->completion
, fd
->comp_key
, req
->cvalue
, req
->status
, req
->information
);
2843 release_object( fd
);
2847 /* set fd completion information */
2848 DECL_HANDLER(set_fd_completion_mode
)
2850 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
2853 if (is_fd_overlapped( fd
))
2855 if (req
->flags
& FILE_SKIP_SET_EVENT_ON_HANDLE
)
2856 set_fd_signaled( fd
, 0 );
2857 /* removing flags is not allowed */
2858 fd
->comp_flags
|= req
->flags
& ( FILE_SKIP_COMPLETION_PORT_ON_SUCCESS
2859 | FILE_SKIP_SET_EVENT_ON_HANDLE
2860 | FILE_SKIP_SET_USER_EVENT_ON_FAST_IO
);
2863 set_error( STATUS_INVALID_PARAMETER
);
2864 release_object( fd
);
2868 /* set fd disposition information */
2869 DECL_HANDLER(set_fd_disp_info
)
2871 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, DELETE
);
2874 set_fd_disposition( fd
, req
->unlink
);
2875 release_object( fd
);
2879 /* set fd name information */
2880 DECL_HANDLER(set_fd_name_info
)
2882 struct fd
*fd
, *root_fd
= NULL
;
2883 struct unicode_str nt_name
;
2885 if (req
->namelen
> get_req_data_size())
2887 set_error( STATUS_INVALID_PARAMETER
);
2890 nt_name
.str
= get_req_data();
2891 nt_name
.len
= (req
->namelen
/ sizeof(WCHAR
)) * sizeof(WCHAR
);
2897 if (!(root
= get_dir_obj( current
->process
, req
->rootdir
, 0 ))) return;
2898 root_fd
= get_obj_fd( (struct object
*)root
);
2899 release_object( root
);
2900 if (!root_fd
) return;
2903 if ((fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 )))
2905 set_fd_name( fd
, root_fd
, (const char *)get_req_data() + req
->namelen
,
2906 get_req_data_size() - req
->namelen
, nt_name
, req
->link
, req
->replace
);
2907 release_object( fd
);
2909 if (root_fd
) release_object( root_fd
);