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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
33 #ifdef HAVE_SYS_POLL_H
41 #include <sys/types.h>
54 #if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
55 # include <sys/epoll.h>
57 #elif defined(linux) && defined(__i386__) && defined(HAVE_STDINT_H)
59 # define EPOLLIN POLLIN
60 # define EPOLLOUT POLLOUT
61 # define EPOLLERR POLLERR
62 # define EPOLLHUP POLLHUP
63 # define EPOLL_CTL_ADD 1
64 # define EPOLL_CTL_DEL 2
65 # define EPOLL_CTL_MOD 3
67 typedef union epoll_data
81 #define SYSCALL_RET(ret) do { \
82 if (ret < 0) { errno = -ret; ret = -1; } \
86 static inline int epoll_create( int size
)
89 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
90 : "=a" (ret
) : "0" (254 /*NR_epoll_create*/), "r" (size
) );
94 static inline int epoll_ctl( int epfd
, int op
, int fd
, const struct epoll_event
*event
)
97 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
99 : "0" (255 /*NR_epoll_ctl*/), "r" (epfd
), "c" (op
), "d" (fd
), "S" (event
), "m" (*event
) );
103 static inline int epoll_wait( int epfd
, struct epoll_event
*events
, int maxevents
, int timeout
)
106 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
108 : "0" (256 /*NR_epoll_wait*/), "r" (epfd
), "c" (events
), "d" (maxevents
), "S" (timeout
)
114 #endif /* linux && __i386__ && HAVE_STDINT_H */
117 /* Because of the stupid Posix locking semantics, we need to keep
118 * track of all file descriptors referencing a given file, and not
119 * close a single one until all the locks are gone (sigh).
122 /* file descriptor object */
124 /* closed_fd is used to keep track of the unix fd belonging to a closed fd object */
127 struct list entry
; /* entry in inode closed list */
128 int fd
; /* the unix file descriptor */
129 char unlink
[1]; /* name to unlink on close (if any) */
134 struct object obj
; /* object header */
135 const struct fd_ops
*fd_ops
; /* file descriptor operations */
136 struct inode
*inode
; /* inode that this fd belongs to */
137 struct list inode_entry
; /* entry in inode fd list */
138 struct closed_fd
*closed
; /* structure to store the unix fd at destroy time */
139 struct object
*user
; /* object using this file descriptor */
140 struct list locks
; /* list of locks on this fd */
141 unsigned int access
; /* file access (GENERIC_READ/WRITE) */
142 unsigned int sharing
; /* file sharing mode */
143 int unix_fd
; /* unix file descriptor */
144 int fs_locks
; /* can we use filesystem locks for this fd? */
145 int poll_index
; /* index of fd in poll array */
148 static void fd_dump( struct object
*obj
, int verbose
);
149 static void fd_destroy( struct object
*obj
);
151 static const struct object_ops fd_ops
=
153 sizeof(struct fd
), /* size */
155 no_add_queue
, /* add_queue */
156 NULL
, /* remove_queue */
158 NULL
, /* satisfied */
159 no_get_fd
, /* get_fd */
160 fd_destroy
/* destroy */
167 struct object obj
; /* object header */
168 struct list entry
; /* inode hash list entry */
169 unsigned int hash
; /* hashing code */
170 dev_t dev
; /* device number */
171 ino_t ino
; /* inode number */
172 struct list open
; /* list of open file descriptors */
173 struct list locks
; /* list of file locks */
174 struct list closed
; /* list of file descriptors to close at destroy time */
177 static void inode_dump( struct object
*obj
, int verbose
);
178 static void inode_destroy( struct object
*obj
);
180 static const struct object_ops inode_ops
=
182 sizeof(struct inode
), /* size */
183 inode_dump
, /* dump */
184 no_add_queue
, /* add_queue */
185 NULL
, /* remove_queue */
187 NULL
, /* satisfied */
188 no_get_fd
, /* get_fd */
189 inode_destroy
/* destroy */
192 /* file lock object */
196 struct object obj
; /* object header */
197 struct fd
*fd
; /* fd owning this lock */
198 struct list fd_entry
; /* entry in list of locks on a given fd */
199 struct list inode_entry
; /* entry in inode list of locks */
200 int shared
; /* shared lock? */
201 file_pos_t start
; /* locked region is interval [start;end) */
203 struct process
*process
; /* process owning this lock */
204 struct list proc_entry
; /* entry in list of locks owned by the process */
207 static void file_lock_dump( struct object
*obj
, int verbose
);
208 static int file_lock_signaled( struct object
*obj
, struct thread
*thread
);
210 static const struct object_ops file_lock_ops
=
212 sizeof(struct file_lock
), /* size */
213 file_lock_dump
, /* dump */
214 add_queue
, /* add_queue */
215 remove_queue
, /* remove_queue */
216 file_lock_signaled
, /* signaled */
217 no_satisfied
, /* satisfied */
218 no_get_fd
, /* get_fd */
219 no_destroy
/* destroy */
223 #define OFF_T_MAX (~((file_pos_t)1 << (8*sizeof(off_t)-1)))
224 #define FILE_POS_T_MAX (~(file_pos_t)0)
226 static file_pos_t max_unix_offset
= OFF_T_MAX
;
228 #define DUMP_LONG_LONG(val) do { \
229 if (sizeof(val) > sizeof(unsigned long) && (val) > ~0UL) \
230 fprintf( stderr, "%lx%08lx", (unsigned long)((val) >> 32), (unsigned long)(val) ); \
232 fprintf( stderr, "%lx", (unsigned long)(val) ); \
237 /****************************************************************/
238 /* timeouts support */
242 struct list entry
; /* entry in sorted timeout list */
243 struct timeval when
; /* timeout expiry (absolute time) */
244 timeout_callback callback
; /* callback function */
245 void *private; /* callback private data */
248 static struct list timeout_list
= LIST_INIT(timeout_list
); /* sorted timeouts list */
250 /* add a timeout user */
251 struct timeout_user
*add_timeout_user( struct timeval
*when
, timeout_callback func
, void *private )
253 struct timeout_user
*user
;
256 if (!(user
= mem_alloc( sizeof(*user
) ))) return NULL
;
258 user
->callback
= func
;
259 user
->private = private;
261 /* Now insert it in the linked list */
263 LIST_FOR_EACH( ptr
, &timeout_list
)
265 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
266 if (!time_before( &timeout
->when
, when
)) break;
268 list_add_before( ptr
, &user
->entry
);
272 /* remove a timeout user */
273 void remove_timeout_user( struct timeout_user
*user
)
275 list_remove( &user
->entry
);
279 /* add a timeout in milliseconds to an absolute time */
280 void add_timeout( struct timeval
*when
, int timeout
)
284 long sec
= timeout
/ 1000;
285 if ((when
->tv_usec
+= (timeout
- 1000*sec
) * 1000) >= 1000000)
287 when
->tv_usec
-= 1000000;
295 /****************************************************************/
298 static struct fd
**poll_users
; /* users array */
299 static struct pollfd
*pollfd
; /* poll fd array */
300 static int nb_users
; /* count of array entries actually in use */
301 static int active_users
; /* current number of active users */
302 static int allocated_users
; /* count of allocated entries in the array */
303 static struct fd
**freelist
; /* list of free entries in the array */
308 static struct epoll_event
*epoll_events
;
310 /* set the events that epoll waits for on this fd; helper for set_fd_events */
311 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
)
313 struct epoll_event ev
;
316 if (epoll_fd
== -1) return;
318 if (events
== -1) /* stop waiting on this fd completely */
320 if (pollfd
[user
].fd
== -1) return; /* already removed */
323 else if (pollfd
[user
].fd
== -1)
325 if (pollfd
[user
].events
) return; /* stopped waiting on it, don't restart */
330 if (pollfd
[user
].events
== events
) return; /* nothing to do */
337 if (epoll_ctl( epoll_fd
, ctl
, fd
->unix_fd
, &ev
) == -1)
339 if (errno
== ENOMEM
) /* not enough memory, give up on epoll */
344 else perror( "epoll_ctl" ); /* should not happen */
348 #else /* USE_EPOLL */
350 static inline void set_fd_epoll_events( struct fd
*fd
, int user
, int events
)
354 #endif /* USE_EPOLL */
357 /* add a user in the poll array and return its index, or -1 on failure */
358 static int add_poll_user( struct fd
*fd
)
363 ret
= freelist
- poll_users
;
364 freelist
= (struct fd
**)poll_users
[ret
];
368 if (nb_users
== allocated_users
)
370 struct fd
**newusers
;
371 struct pollfd
*newpoll
;
372 int new_count
= allocated_users
? (allocated_users
+ allocated_users
/ 2) : 16;
373 if (!(newusers
= realloc( poll_users
, new_count
* sizeof(*poll_users
) ))) return -1;
374 if (!(newpoll
= realloc( pollfd
, new_count
* sizeof(*pollfd
) )))
377 poll_users
= newusers
;
382 poll_users
= newusers
;
385 if (!allocated_users
) epoll_fd
= epoll_create( new_count
);
388 struct epoll_event
*new_events
;
389 if (!(new_events
= realloc( epoll_events
, new_count
* sizeof(*epoll_events
) )))
391 epoll_events
= new_events
;
394 allocated_users
= new_count
;
399 pollfd
[ret
].events
= 0;
400 pollfd
[ret
].revents
= 0;
401 poll_users
[ret
] = fd
;
406 /* remove a user from the poll list */
407 static void remove_poll_user( struct fd
*fd
, int user
)
410 assert( poll_users
[user
] == fd
);
413 if (epoll_fd
!= -1 && pollfd
[user
].fd
!= -1)
415 struct epoll_event dummy
;
416 epoll_ctl( epoll_fd
, EPOLL_CTL_DEL
, fd
->unix_fd
, &dummy
);
419 pollfd
[user
].fd
= -1;
420 pollfd
[user
].events
= 0;
421 pollfd
[user
].revents
= 0;
422 poll_users
[user
] = (struct fd
*)freelist
;
423 freelist
= &poll_users
[user
];
427 /* process pending timeouts and return the time until the next timeout, in milliseconds */
428 static int get_next_timeout(void)
430 if (!list_empty( &timeout_list
))
432 struct list expired_list
, *ptr
;
435 gettimeofday( &now
, NULL
);
437 /* first remove all expired timers from the list */
439 list_init( &expired_list
);
440 while ((ptr
= list_head( &timeout_list
)) != NULL
)
442 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
444 if (!time_before( &now
, &timeout
->when
))
446 list_remove( &timeout
->entry
);
447 list_add_tail( &expired_list
, &timeout
->entry
);
452 /* now call the callback for all the removed timers */
454 while ((ptr
= list_head( &expired_list
)) != NULL
)
456 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
457 list_remove( &timeout
->entry
);
458 timeout
->callback( timeout
->private );
462 if ((ptr
= list_head( &timeout_list
)) != NULL
)
464 struct timeout_user
*timeout
= LIST_ENTRY( ptr
, struct timeout_user
, entry
);
465 int diff
= (timeout
->when
.tv_sec
- now
.tv_sec
) * 1000
466 + (timeout
->when
.tv_usec
- now
.tv_usec
) / 1000;
467 if (diff
< 0) diff
= 0;
471 return -1; /* no pending timeouts */
474 /* server main poll() loop */
480 assert( POLLIN
== EPOLLIN
);
481 assert( POLLOUT
== EPOLLOUT
);
482 assert( POLLERR
== EPOLLERR
);
483 assert( POLLHUP
== EPOLLHUP
);
489 timeout
= get_next_timeout();
491 if (!active_users
) break; /* last user removed by a timeout */
492 if (epoll_fd
== -1) break; /* an error occurred with epoll */
494 ret
= epoll_wait( epoll_fd
, epoll_events
, allocated_users
, timeout
);
496 /* put the events into the pollfd array first, like poll does */
497 for (i
= 0; i
< ret
; i
++)
499 int user
= epoll_events
[i
].data
.u32
;
500 pollfd
[user
].revents
= epoll_events
[i
].events
;
503 /* read events from the pollfd array, as set_fd_events may modify them */
504 for (i
= 0; i
< ret
; i
++)
506 int user
= epoll_events
[i
].data
.u32
;
507 if (pollfd
[user
].revents
) fd_poll_event( poll_users
[user
], pollfd
[user
].revents
);
511 /* fall through to normal poll loop */
512 #endif /* USE_EPOLL */
516 timeout
= get_next_timeout();
518 if (!active_users
) break; /* last user removed by a timeout */
520 ret
= poll( pollfd
, nb_users
, timeout
);
523 for (i
= 0; i
< nb_users
; i
++)
525 if (pollfd
[i
].revents
)
527 fd_poll_event( poll_users
[i
], pollfd
[i
].revents
);
536 /****************************************************************/
537 /* inode functions */
541 static struct list inode_hash
[HASH_SIZE
];
543 /* close all pending file descriptors in the closed list */
544 static void inode_close_pending( struct inode
*inode
)
546 struct list
*ptr
= list_head( &inode
->closed
);
550 struct closed_fd
*fd
= LIST_ENTRY( ptr
, struct closed_fd
, entry
);
551 struct list
*next
= list_next( &inode
->closed
, ptr
);
558 if (!fd
->unlink
) /* get rid of it unless there's an unlink pending on that file */
568 static void inode_dump( struct object
*obj
, int verbose
)
570 struct inode
*inode
= (struct inode
*)obj
;
571 fprintf( stderr
, "Inode dev=" );
572 DUMP_LONG_LONG( inode
->dev
);
573 fprintf( stderr
, " ino=" );
574 DUMP_LONG_LONG( inode
->ino
);
575 fprintf( stderr
, "\n" );
578 static void inode_destroy( struct object
*obj
)
580 struct inode
*inode
= (struct inode
*)obj
;
583 assert( list_empty(&inode
->open
) );
584 assert( list_empty(&inode
->locks
) );
586 list_remove( &inode
->entry
);
588 while ((ptr
= list_head( &inode
->closed
)))
590 struct closed_fd
*fd
= LIST_ENTRY( ptr
, struct closed_fd
, entry
);
592 if (fd
->fd
!= -1) close( fd
->fd
);
595 /* make sure it is still the same file */
597 if (!stat( fd
->unlink
, &st
) && st
.st_dev
== inode
->dev
&& st
.st_ino
== inode
->ino
)
599 if (S_ISDIR(st
.st_mode
)) rmdir( fd
->unlink
);
600 else unlink( fd
->unlink
);
607 /* retrieve the inode object for a given fd, creating it if needed */
608 static struct inode
*get_inode( dev_t dev
, ino_t ino
)
612 unsigned int hash
= (dev
^ ino
) % HASH_SIZE
;
614 if (inode_hash
[hash
].next
)
616 LIST_FOR_EACH( ptr
, &inode_hash
[hash
] )
618 inode
= LIST_ENTRY( ptr
, struct inode
, entry
);
619 if (inode
->dev
== dev
&& inode
->ino
== ino
)
620 return (struct inode
*)grab_object( inode
);
623 else list_init( &inode_hash
[hash
] );
625 /* not found, create it */
626 if ((inode
= alloc_object( &inode_ops
)))
631 list_init( &inode
->open
);
632 list_init( &inode
->locks
);
633 list_init( &inode
->closed
);
634 list_add_head( &inode_hash
[hash
], &inode
->entry
);
639 /* add fd to the indoe list of file descriptors to close */
640 static void inode_add_closed_fd( struct inode
*inode
, struct closed_fd
*fd
)
642 if (!list_empty( &inode
->locks
))
644 list_add_head( &inode
->closed
, &fd
->entry
);
646 else if (fd
->unlink
[0]) /* close the fd but keep the structure around for unlink */
650 list_add_head( &inode
->closed
, &fd
->entry
);
652 else /* no locks on this inode and no unlink, get rid of the fd */
660 /****************************************************************/
661 /* file lock functions */
663 static void file_lock_dump( struct object
*obj
, int verbose
)
665 struct file_lock
*lock
= (struct file_lock
*)obj
;
666 fprintf( stderr
, "Lock %s fd=%p proc=%p start=",
667 lock
->shared
? "shared" : "excl", lock
->fd
, lock
->process
);
668 DUMP_LONG_LONG( lock
->start
);
669 fprintf( stderr
, " end=" );
670 DUMP_LONG_LONG( lock
->end
);
671 fprintf( stderr
, "\n" );
674 static int file_lock_signaled( struct object
*obj
, struct thread
*thread
)
676 struct file_lock
*lock
= (struct file_lock
*)obj
;
677 /* lock is signaled if it has lost its owner */
678 return !lock
->process
;
681 /* set (or remove) a Unix lock if possible for the given range */
682 static int set_unix_lock( struct fd
*fd
, file_pos_t start
, file_pos_t end
, int type
)
686 if (!fd
->fs_locks
) return 1; /* no fs locks possible for this fd */
689 if (start
== end
) return 1; /* can't set zero-byte lock */
690 if (start
> max_unix_offset
) return 1; /* ignore it */
692 fl
.l_whence
= SEEK_SET
;
694 if (!end
|| end
> max_unix_offset
) fl
.l_len
= 0;
695 else fl
.l_len
= end
- start
;
696 if (fcntl( fd
->unix_fd
, F_SETLK
, &fl
) != -1) return 1;
701 /* check whether locks work at all on this file system */
702 if (fcntl( fd
->unix_fd
, F_GETLK
, &fl
) != -1)
704 set_error( STATUS_FILE_LOCK_CONFLICT
);
710 /* no locking on this fs, just ignore it */
714 set_error( STATUS_FILE_LOCK_CONFLICT
);
717 /* this can happen if we try to set a write lock on a read-only file */
718 /* we just ignore that error */
719 if (fl
.l_type
== F_WRLCK
) return 1;
720 set_error( STATUS_ACCESS_DENIED
);
726 /* this can happen if off_t is 64-bit but the kernel only supports 32-bit */
727 /* in that case we shrink the limit and retry */
728 if (max_unix_offset
> INT_MAX
)
730 max_unix_offset
= INT_MAX
;
741 /* check if interval [start;end) overlaps the lock */
742 inline static int lock_overlaps( struct file_lock
*lock
, file_pos_t start
, file_pos_t end
)
744 if (lock
->end
&& start
>= lock
->end
) return 0;
745 if (end
&& lock
->start
>= end
) return 0;
749 /* remove Unix locks for all bytes in the specified area that are no longer locked */
750 static void remove_unix_locks( struct fd
*fd
, file_pos_t start
, file_pos_t end
)
758 } *first
, *cur
, *next
, *buffer
;
763 if (!fd
->inode
) return;
764 if (!fd
->fs_locks
) return;
765 if (start
== end
|| start
> max_unix_offset
) return;
766 if (!end
|| end
> max_unix_offset
) end
= max_unix_offset
+ 1;
768 /* count the number of locks overlapping the specified area */
770 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
772 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
773 if (lock
->start
== lock
->end
) continue;
774 if (lock_overlaps( lock
, start
, end
)) count
++;
777 if (!count
) /* no locks at all, we can unlock everything */
779 set_unix_lock( fd
, start
, end
, F_UNLCK
);
783 /* allocate space for the list of holes */
784 /* max. number of holes is number of locks + 1 */
786 if (!(buffer
= malloc( sizeof(*buffer
) * (count
+1) ))) return;
790 first
->start
= start
;
794 /* build a sorted list of unlocked holes in the specified area */
796 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
798 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
799 if (lock
->start
== lock
->end
) continue;
800 if (!lock_overlaps( lock
, start
, end
)) continue;
802 /* go through all the holes touched by this lock */
803 for (cur
= first
; cur
; cur
= cur
->next
)
805 if (cur
->end
<= lock
->start
) continue; /* hole is before start of lock */
806 if (lock
->end
&& cur
->start
>= lock
->end
) break; /* hole is after end of lock */
808 /* now we know that lock is overlapping hole */
810 if (cur
->start
>= lock
->start
) /* lock starts before hole, shrink from start */
812 cur
->start
= lock
->end
;
813 if (cur
->start
&& cur
->start
< cur
->end
) break; /* done with this lock */
814 /* now hole is empty, remove it */
815 if (cur
->next
) cur
->next
->prev
= cur
->prev
;
816 if (cur
->prev
) cur
->prev
->next
= cur
->next
;
817 else if (!(first
= cur
->next
)) goto done
; /* no more holes at all */
819 else if (!lock
->end
|| cur
->end
<= lock
->end
) /* lock larger than hole, shrink from end */
821 cur
->end
= lock
->start
;
822 assert( cur
->start
< cur
->end
);
824 else /* lock is in the middle of hole, split hole in two */
827 next
->next
= cur
->next
;
829 next
->start
= lock
->end
;
830 next
->end
= cur
->end
;
831 cur
->end
= lock
->start
;
832 assert( next
->start
< next
->end
);
833 assert( cur
->end
< next
->start
);
835 break; /* done with this lock */
840 /* clear Unix locks for all the holes */
842 for (cur
= first
; cur
; cur
= cur
->next
)
843 set_unix_lock( fd
, cur
->start
, cur
->end
, F_UNLCK
);
849 /* create a new lock on a fd */
850 static struct file_lock
*add_lock( struct fd
*fd
, int shared
, file_pos_t start
, file_pos_t end
)
852 struct file_lock
*lock
;
854 if (!fd
->inode
) /* not a regular file */
856 set_error( STATUS_INVALID_HANDLE
);
860 if (!(lock
= alloc_object( &file_lock_ops
))) return NULL
;
861 lock
->shared
= shared
;
865 lock
->process
= current
->process
;
867 /* now try to set a Unix lock */
868 if (!set_unix_lock( lock
->fd
, lock
->start
, lock
->end
, lock
->shared
? F_RDLCK
: F_WRLCK
))
870 release_object( lock
);
873 list_add_head( &fd
->locks
, &lock
->fd_entry
);
874 list_add_head( &fd
->inode
->locks
, &lock
->inode_entry
);
875 list_add_head( &lock
->process
->locks
, &lock
->proc_entry
);
879 /* remove an existing lock */
880 static void remove_lock( struct file_lock
*lock
, int remove_unix
)
882 struct inode
*inode
= lock
->fd
->inode
;
884 list_remove( &lock
->fd_entry
);
885 list_remove( &lock
->inode_entry
);
886 list_remove( &lock
->proc_entry
);
887 if (remove_unix
) remove_unix_locks( lock
->fd
, lock
->start
, lock
->end
);
888 if (list_empty( &inode
->locks
)) inode_close_pending( inode
);
889 lock
->process
= NULL
;
890 wake_up( &lock
->obj
, 0 );
891 release_object( lock
);
894 /* remove all locks owned by a given process */
895 void remove_process_locks( struct process
*process
)
899 while ((ptr
= list_head( &process
->locks
)))
901 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, proc_entry
);
902 remove_lock( lock
, 1 ); /* this removes it from the list */
906 /* remove all locks on a given fd */
907 static void remove_fd_locks( struct fd
*fd
)
909 file_pos_t start
= FILE_POS_T_MAX
, end
= 0;
912 while ((ptr
= list_head( &fd
->locks
)))
914 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, fd_entry
);
915 if (lock
->start
< start
) start
= lock
->start
;
916 if (!lock
->end
|| lock
->end
> end
) end
= lock
->end
- 1;
917 remove_lock( lock
, 0 );
919 if (start
< end
) remove_unix_locks( fd
, start
, end
+ 1 );
922 /* add a lock on an fd */
923 /* returns handle to wait on */
924 obj_handle_t
lock_fd( struct fd
*fd
, file_pos_t start
, file_pos_t count
, int shared
, int wait
)
927 file_pos_t end
= start
+ count
;
929 /* don't allow wrapping locks */
930 if (end
&& end
< start
)
932 set_error( STATUS_INVALID_PARAMETER
);
936 /* check if another lock on that file overlaps the area */
937 LIST_FOR_EACH( ptr
, &fd
->inode
->locks
)
939 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, inode_entry
);
940 if (!lock_overlaps( lock
, start
, end
)) continue;
941 if (lock
->shared
&& shared
) continue;
945 set_error( STATUS_FILE_LOCK_CONFLICT
);
948 set_error( STATUS_PENDING
);
949 return alloc_handle( current
->process
, lock
, SYNCHRONIZE
, 0 );
952 /* not found, add it */
953 if (add_lock( fd
, shared
, start
, end
)) return 0;
954 if (get_error() == STATUS_FILE_LOCK_CONFLICT
)
956 /* Unix lock conflict -> tell client to wait and retry */
957 if (wait
) set_error( STATUS_PENDING
);
962 /* remove a lock on an fd */
963 void unlock_fd( struct fd
*fd
, file_pos_t start
, file_pos_t count
)
966 file_pos_t end
= start
+ count
;
968 /* find an existing lock with the exact same parameters */
969 LIST_FOR_EACH( ptr
, &fd
->locks
)
971 struct file_lock
*lock
= LIST_ENTRY( ptr
, struct file_lock
, fd_entry
);
972 if ((lock
->start
== start
) && (lock
->end
== end
))
974 remove_lock( lock
, 1 );
978 set_error( STATUS_FILE_LOCK_CONFLICT
);
982 /****************************************************************/
983 /* asynchronous operations support */
988 struct thread
*thread
;
993 struct timeout_user
*timeout
;
997 /* notifies client thread of new status of its async request */
998 /* destroys the server side of it */
999 static void async_terminate( struct async
*async
, int status
)
1001 thread_queue_apc( async
->thread
, NULL
, async
->apc
, APC_ASYNC_IO
,
1002 1, async
->user
, async
->sb
, (void *)status
);
1004 if (async
->timeout
) remove_timeout_user( async
->timeout
);
1005 async
->timeout
= NULL
;
1006 list_remove( &async
->entry
);
1007 release_object( async
->thread
);
1011 /* cb for timeout on an async request */
1012 static void async_callback(void *private)
1014 struct async
*async
= (struct async
*)private;
1016 /* fprintf(stderr, "async timeout out %p\n", async); */
1017 async
->timeout
= NULL
;
1018 async_terminate( async
, STATUS_TIMEOUT
);
1021 /* create an async on a given queue of a fd */
1022 struct async
*create_async(struct fd
*fd
, struct thread
*thread
, int timeout
, struct list
*queue
,
1023 void *io_apc
, void *io_user
, void* io_sb
)
1025 struct async
*async
= mem_alloc( sizeof(struct async
) );
1027 if (!async
) return NULL
;
1030 async
->thread
= (struct thread
*)grab_object(thread
);
1031 async
->apc
= io_apc
;
1032 async
->user
= io_user
;
1035 list_add_tail( queue
, &async
->entry
);
1039 gettimeofday( &async
->when
, 0 );
1040 add_timeout( &async
->when
, timeout
);
1041 async
->timeout
= add_timeout_user( &async
->when
, async_callback
, async
);
1043 else async
->timeout
= NULL
;
1048 /* terminate the async operation at the head of the queue */
1049 void async_terminate_head( struct list
*queue
, int status
)
1051 struct list
*ptr
= list_head( queue
);
1052 if (ptr
) async_terminate( LIST_ENTRY( ptr
, struct async
, entry
), status
);
1055 /****************************************************************/
1056 /* file descriptor functions */
1058 static void fd_dump( struct object
*obj
, int verbose
)
1060 struct fd
*fd
= (struct fd
*)obj
;
1061 fprintf( stderr
, "Fd unix_fd=%d user=%p", fd
->unix_fd
, fd
->user
);
1062 if (fd
->inode
) fprintf( stderr
, " inode=%p unlink='%s'", fd
->inode
, fd
->closed
->unlink
);
1063 fprintf( stderr
, "\n" );
1066 static void fd_destroy( struct object
*obj
)
1068 struct fd
*fd
= (struct fd
*)obj
;
1070 remove_fd_locks( fd
);
1071 list_remove( &fd
->inode_entry
);
1072 if (fd
->poll_index
!= -1) remove_poll_user( fd
, fd
->poll_index
);
1075 inode_add_closed_fd( fd
->inode
, fd
->closed
);
1076 release_object( fd
->inode
);
1078 else /* no inode, close it right away */
1080 if (fd
->unix_fd
!= -1) close( fd
->unix_fd
);
1084 /* set the events that select waits for on this fd */
1085 void set_fd_events( struct fd
*fd
, int events
)
1087 int user
= fd
->poll_index
;
1088 assert( poll_users
[user
] == fd
);
1090 set_fd_epoll_events( fd
, user
, events
);
1092 if (events
== -1) /* stop waiting on this fd completely */
1094 pollfd
[user
].fd
= -1;
1095 pollfd
[user
].events
= POLLERR
;
1096 pollfd
[user
].revents
= 0;
1098 else if (pollfd
[user
].fd
!= -1 || !pollfd
[user
].events
)
1100 pollfd
[user
].fd
= fd
->unix_fd
;
1101 pollfd
[user
].events
= events
;
1105 /* allocate an fd object, without setting the unix fd yet */
1106 struct fd
*alloc_fd( const struct fd_ops
*fd_user_ops
, struct object
*user
)
1108 struct fd
*fd
= alloc_object( &fd_ops
);
1110 if (!fd
) return NULL
;
1112 fd
->fd_ops
= fd_user_ops
;
1120 fd
->poll_index
= -1;
1121 list_init( &fd
->inode_entry
);
1122 list_init( &fd
->locks
);
1124 if ((fd
->poll_index
= add_poll_user( fd
)) == -1)
1126 release_object( fd
);
1132 /* check if the desired access is possible without violating */
1133 /* the sharing mode of other opens of the same file */
1134 static int check_sharing( struct fd
*fd
, unsigned int access
, unsigned int sharing
)
1136 unsigned int existing_sharing
= FILE_SHARE_READ
| FILE_SHARE_WRITE
| FILE_SHARE_DELETE
;
1137 unsigned int existing_access
= 0;
1141 /* if access mode is 0, sharing mode is ignored */
1142 if (!access
) sharing
= existing_sharing
;
1143 fd
->access
= access
;
1144 fd
->sharing
= sharing
;
1146 LIST_FOR_EACH( ptr
, &fd
->inode
->open
)
1148 struct fd
*fd_ptr
= LIST_ENTRY( ptr
, struct fd
, inode_entry
);
1151 existing_sharing
&= fd_ptr
->sharing
;
1152 existing_access
|= fd_ptr
->access
;
1153 if (fd_ptr
->closed
->unlink
[0]) unlink
= 1;
1157 if ((access
& GENERIC_READ
) && !(existing_sharing
& FILE_SHARE_READ
)) return 0;
1158 if ((access
& GENERIC_WRITE
) && !(existing_sharing
& FILE_SHARE_WRITE
)) return 0;
1159 if ((existing_access
& GENERIC_READ
) && !(sharing
& FILE_SHARE_READ
)) return 0;
1160 if ((existing_access
& GENERIC_WRITE
) && !(sharing
& FILE_SHARE_WRITE
)) return 0;
1161 if (fd
->closed
->unlink
[0] && !(existing_sharing
& FILE_SHARE_DELETE
)) return 0;
1162 if (unlink
&& !(sharing
& FILE_SHARE_DELETE
)) return 0;
1166 /* open() wrapper using a struct fd */
1167 /* the fd must have been created with alloc_fd */
1168 /* on error the fd object is released */
1169 struct fd
*open_fd( struct fd
*fd
, const char *name
, int flags
, mode_t
*mode
,
1170 unsigned int access
, unsigned int sharing
, unsigned int options
)
1173 struct closed_fd
*closed_fd
;
1174 const char *unlink_name
= "";
1176 assert( fd
->unix_fd
== -1 );
1178 if (options
& FILE_DELETE_ON_CLOSE
) unlink_name
= name
;
1179 if (!(closed_fd
= mem_alloc( sizeof(*closed_fd
) + strlen(unlink_name
) )))
1181 release_object( fd
);
1184 /* create the directory if needed */
1185 if ((options
& FILE_DIRECTORY_FILE
) && (flags
& O_CREAT
))
1187 if (mkdir( name
, 0777 ) == -1)
1189 if (errno
!= EEXIST
|| (flags
& O_EXCL
))
1192 release_object( fd
);
1197 flags
&= ~(O_CREAT
| O_EXCL
| O_TRUNC
);
1199 if ((fd
->unix_fd
= open( name
, flags
& ~O_TRUNC
, *mode
)) == -1)
1202 release_object( fd
);
1206 closed_fd
->fd
= fd
->unix_fd
;
1207 closed_fd
->unlink
[0] = 0;
1208 fstat( fd
->unix_fd
, &st
);
1211 /* only bother with an inode for normal files and directories */
1212 if (S_ISREG(st
.st_mode
) || S_ISDIR(st
.st_mode
))
1214 struct inode
*inode
= get_inode( st
.st_dev
, st
.st_ino
);
1218 /* we can close the fd because there are no others open on the same file,
1219 * otherwise we wouldn't have failed to allocate a new inode
1224 fd
->closed
= closed_fd
;
1225 list_add_head( &inode
->open
, &fd
->inode_entry
);
1227 /* check directory options */
1228 if ((options
& FILE_DIRECTORY_FILE
) && !S_ISDIR(st
.st_mode
))
1230 release_object( fd
);
1231 set_error( STATUS_NOT_A_DIRECTORY
);
1234 if ((options
& FILE_NON_DIRECTORY_FILE
) && S_ISDIR(st
.st_mode
))
1236 release_object( fd
);
1237 set_error( STATUS_FILE_IS_A_DIRECTORY
);
1240 if (!check_sharing( fd
, access
, sharing
))
1242 release_object( fd
);
1243 set_error( STATUS_SHARING_VIOLATION
);
1246 strcpy( closed_fd
->unlink
, unlink_name
);
1247 if (flags
& O_TRUNC
) ftruncate( fd
->unix_fd
, 0 );
1249 else /* special file */
1251 if (options
& FILE_DIRECTORY_FILE
)
1253 set_error( STATUS_NOT_A_DIRECTORY
);
1256 if (unlink_name
[0]) /* we can't unlink special files */
1258 set_error( STATUS_INVALID_PARAMETER
);
1266 release_object( fd
);
1271 /* create an fd for an anonymous file */
1272 /* if the function fails the unix fd is closed */
1273 struct fd
*create_anonymous_fd( const struct fd_ops
*fd_user_ops
, int unix_fd
, struct object
*user
)
1275 struct fd
*fd
= alloc_fd( fd_user_ops
, user
);
1279 fd
->unix_fd
= unix_fd
;
1286 /* retrieve the object that is using an fd */
1287 void *get_fd_user( struct fd
*fd
)
1292 /* retrieve the unix fd for an object */
1293 int get_unix_fd( struct fd
*fd
)
1298 /* check if two file descriptors point to the same file */
1299 int is_same_file_fd( struct fd
*fd1
, struct fd
*fd2
)
1301 return fd1
->inode
== fd2
->inode
;
1304 /* callback for event happening in the main poll() loop */
1305 void fd_poll_event( struct fd
*fd
, int event
)
1307 return fd
->fd_ops
->poll_event( fd
, event
);
1310 /* check if events are pending and if yes return which one(s) */
1311 int check_fd_events( struct fd
*fd
, int events
)
1315 pfd
.fd
= fd
->unix_fd
;
1316 pfd
.events
= events
;
1317 if (poll( &pfd
, 1, 0 ) <= 0) return 0;
1321 /* default add_queue() routine for objects that poll() on an fd */
1322 int default_fd_add_queue( struct object
*obj
, struct wait_queue_entry
*entry
)
1324 struct fd
*fd
= get_obj_fd( obj
);
1327 if (!obj
->head
) /* first on the queue */
1328 set_fd_events( fd
, fd
->fd_ops
->get_poll_events( fd
) );
1329 add_queue( obj
, entry
);
1330 release_object( fd
);
1334 /* default remove_queue() routine for objects that poll() on an fd */
1335 void default_fd_remove_queue( struct object
*obj
, struct wait_queue_entry
*entry
)
1337 struct fd
*fd
= get_obj_fd( obj
);
1340 remove_queue( obj
, entry
);
1341 if (!obj
->head
) /* last on the queue is gone */
1342 set_fd_events( fd
, 0 );
1343 release_object( obj
);
1344 release_object( fd
);
1347 /* default signaled() routine for objects that poll() on an fd */
1348 int default_fd_signaled( struct object
*obj
, struct thread
*thread
)
1351 struct fd
*fd
= get_obj_fd( obj
);
1353 if (fd
->inode
) return 1; /* regular files are always signaled */
1355 events
= fd
->fd_ops
->get_poll_events( fd
);
1356 ret
= check_fd_events( fd
, events
) != 0;
1359 set_fd_events( fd
, 0 ); /* stop waiting on select() if we are signaled */
1361 set_fd_events( fd
, events
); /* restart waiting on poll() if we are no longer signaled */
1363 release_object( fd
);
1367 /* default handler for poll() events */
1368 void default_poll_event( struct fd
*fd
, int event
)
1370 /* an error occurred, stop polling this fd to avoid busy-looping */
1371 if (event
& (POLLERR
| POLLHUP
)) set_fd_events( fd
, -1 );
1372 wake_up( fd
->user
, 0 );
1375 /* default flush() routine */
1376 int no_flush( struct fd
*fd
, struct event
**event
)
1378 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
1382 /* default get_file_info() routine */
1383 int no_get_file_info( struct fd
*fd
)
1385 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
1389 /* default queue_async() routine */
1390 void no_queue_async( struct fd
*fd
, void* apc
, void* user
, void* io_sb
,
1391 int type
, int count
)
1393 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
1396 /* default cancel_async() routine */
1397 void no_cancel_async( struct fd
*fd
)
1399 set_error( STATUS_OBJECT_TYPE_MISMATCH
);
1402 /* same as get_handle_obj but retrieve the struct fd associated to the object */
1403 static struct fd
*get_handle_fd_obj( struct process
*process
, obj_handle_t handle
,
1404 unsigned int access
)
1406 struct fd
*fd
= NULL
;
1409 if ((obj
= get_handle_obj( process
, handle
, access
, NULL
)))
1411 fd
= get_obj_fd( obj
);
1412 release_object( obj
);
1417 /* flush a file buffers */
1418 DECL_HANDLER(flush_file
)
1420 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
1421 struct event
* event
= NULL
;
1425 fd
->fd_ops
->flush( fd
, &event
);
1428 reply
->event
= alloc_handle( current
->process
, event
, SYNCHRONIZE
, 0 );
1430 release_object( fd
);
1434 /* get a Unix fd to access a file */
1435 DECL_HANDLER(get_handle_fd
)
1441 if ((fd
= get_handle_fd_obj( current
->process
, req
->handle
, req
->access
)))
1443 int unix_fd
= get_handle_unix_fd( current
->process
, req
->handle
, req
->access
);
1444 if (unix_fd
!= -1) reply
->fd
= unix_fd
;
1445 else if (!get_error())
1447 assert( fd
->unix_fd
!= -1 );
1448 send_client_fd( current
->process
, fd
->unix_fd
, req
->handle
);
1450 reply
->flags
= fd
->fd_ops
->get_file_info( fd
);
1451 release_object( fd
);
1455 /* create / reschedule an async I/O */
1456 DECL_HANDLER(register_async
)
1458 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
1461 * The queue_async method must do the following:
1463 * 1. Get the async_queue for the request of given type.
1464 * 2. Create a new asynchronous request for the selected queue
1465 * 3. Carry out any operations necessary to adjust the object's poll events
1466 * Usually: set_elect_events (obj, obj->ops->get_poll_events()).
1467 * 4. When the async request is triggered, then send back (with a proper APC)
1468 * the trigger (STATUS_ALERTED) to the thread that posted the request.
1469 * async_destroy() is to be called: it will both notify the sender about
1470 * the trigger and destroy the request by itself
1471 * See also the implementations in file.c, serial.c, and sock.c.
1476 fd
->fd_ops
->queue_async( fd
, req
->io_apc
, req
->io_user
, req
->io_sb
,
1477 req
->type
, req
->count
);
1478 release_object( fd
);
1482 /* cancels all async I/O */
1483 DECL_HANDLER(cancel_async
)
1485 struct fd
*fd
= get_handle_fd_obj( current
->process
, req
->handle
, 0 );
1488 /* Note: we don't kill the queued APC_ASYNC_IO on this thread because
1489 * NtCancelIoFile() will force the pending APC to be run. Since,
1490 * Windows only guarantees that the current thread will have no async
1491 * operation on the current fd when NtCancelIoFile returns, this shall
1494 fd
->fd_ops
->cancel_async( fd
);
1495 release_object( fd
);