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
36 #ifdef HAVE_SYS_POLL_H
44 #include <sys/types.h>
55 #if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
56 # include <sys/epoll.h>
58 #elif defined(linux) && defined(__i386__) && defined(HAVE_STDINT_H)
60 # define EPOLLIN POLLIN
61 # define EPOLLOUT POLLOUT
62 # define EPOLLERR POLLERR
63 # define EPOLLHUP POLLHUP
64 # define EPOLL_CTL_ADD 1
65 # define EPOLL_CTL_DEL 2
66 # define EPOLL_CTL_MOD 3
68 typedef union epoll_data
82 #define SYSCALL_RET(ret) do { \
83 if (ret < 0) { errno = -ret; ret = -1; } \
87 static inline int epoll_create( int size )
90 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
91 : "=a" (ret) : "0" (254 /*NR_epoll_create*/), "r" (size) );
95 static inline int epoll_ctl( int epfd, int op, int fd, const struct epoll_event *event )
98 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
100 : "0" (255 /*NR_epoll_ctl*/), "r" (epfd), "c" (op), "d" (fd), "S" (event), "m" (*event) );
104 static inline int epoll_wait( int epfd, struct epoll_event *events, int maxevents, int timeout )
107 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
109 : "0" (256 /*NR_epoll_wait*/), "r" (epfd), "c" (events), "d" (maxevents), "S" (timeout)
115 #endif /* linux && __i386__ && HAVE_STDINT_H */
118 /* Because of the stupid Posix locking semantics, we need to keep
119 * track of all file descriptors referencing a given file, and not
120 * close a single one until all the locks are gone (sigh).
123 /* file descriptor object */
125 /* closed_fd is used to keep track of the unix fd belonging to a closed fd object */
128 struct list entry; /* entry in inode closed list */
129 int fd; /* the unix file descriptor */
130 char unlink[1]; /* name to unlink on close (if any) */
135 struct object obj; /* object header */
136 const struct fd_ops *fd_ops; /* file descriptor operations */
137 struct inode *inode; /* inode that this fd belongs to */
138 struct list inode_entry; /* entry in inode fd list */
139 struct closed_fd *closed; /* structure to store the unix fd at destroy time */
140 struct object *user; /* object using this file descriptor */
141 struct list locks; /* list of locks on this fd */
142 unsigned int access; /* file access (GENERIC_READ/WRITE) */
143 unsigned int sharing; /* file sharing mode */
144 int unix_fd; /* unix file descriptor */
145 int fs_locks; /* can we use filesystem locks for this fd? */
146 int poll_index; /* index of fd in poll array */
147 struct list read_q; /* async readers of this fd */
148 struct list write_q; /* async writers of this fd */
151 static void fd_dump( struct object *obj, int verbose );
152 static void fd_destroy( struct object *obj );
154 static const struct object_ops fd_ops =
156 sizeof(struct fd), /* size */
158 no_add_queue, /* add_queue */
159 NULL, /* remove_queue */
161 NULL, /* satisfied */
162 no_signal, /* signal */
163 no_get_fd, /* get_fd */
164 no_close_handle, /* close_handle */
165 fd_destroy /* destroy */
172 struct object obj; /* object header */
173 struct list entry; /* inode hash list entry */
174 unsigned int hash; /* hashing code */
175 dev_t dev; /* device number */
176 ino_t ino; /* inode number */
177 struct list open; /* list of open file descriptors */
178 struct list locks; /* list of file locks */
179 struct list closed; /* list of file descriptors to close at destroy time */
182 static void inode_dump( struct object *obj, int verbose );
183 static void inode_destroy( struct object *obj );
185 static const struct object_ops inode_ops =
187 sizeof(struct inode), /* size */
188 inode_dump, /* dump */
189 no_add_queue, /* add_queue */
190 NULL, /* remove_queue */
192 NULL, /* satisfied */
193 no_signal, /* signal */
194 no_get_fd, /* get_fd */
195 no_close_handle, /* close_handle */
196 inode_destroy /* destroy */
199 /* file lock object */
203 struct object obj; /* object header */
204 struct fd *fd; /* fd owning this lock */
205 struct list fd_entry; /* entry in list of locks on a given fd */
206 struct list inode_entry; /* entry in inode list of locks */
207 int shared; /* shared lock? */
208 file_pos_t start; /* locked region is interval [start;end) */
210 struct process *process; /* process owning this lock */
211 struct list proc_entry; /* entry in list of locks owned by the process */
214 static void file_lock_dump( struct object *obj, int verbose );
215 static int file_lock_signaled( struct object *obj, struct thread *thread );
217 static const struct object_ops file_lock_ops =
219 sizeof(struct file_lock), /* size */
220 file_lock_dump, /* dump */
221 add_queue, /* add_queue */
222 remove_queue, /* remove_queue */
223 file_lock_signaled, /* signaled */
224 no_satisfied, /* satisfied */
225 no_signal, /* signal */
226 no_get_fd, /* get_fd */
227 no_close_handle, /* close_handle */
228 no_destroy /* destroy */
232 #define OFF_T_MAX (~((file_pos_t)1 << (8*sizeof(off_t)-1)))
233 #define FILE_POS_T_MAX (~(file_pos_t)0)
235 static file_pos_t max_unix_offset = OFF_T_MAX;
237 #define DUMP_LONG_LONG(val) do { \
238 if (sizeof(val) > sizeof(unsigned long) && (val) > ~0UL) \
239 fprintf( stderr, "%lx%08lx", (unsigned long)((val) >> 32), (unsigned long)(val) ); \
241 fprintf( stderr, "%lx", (unsigned long)(val) ); \
246 /****************************************************************/
247 /* timeouts support */
251 struct list entry; /* entry in sorted timeout list */
252 struct timeval when; /* timeout expiry (absolute time) */
253 timeout_callback callback; /* callback function */
254 void *private; /* callback private data */
257 static struct list timeout_list = LIST_INIT(timeout_list); /* sorted timeouts list */
259 /* add a timeout user */
260 struct timeout_user *add_timeout_user( const struct timeval *when, timeout_callback func,
263 struct timeout_user *user;
266 if (!(user = mem_alloc( sizeof(*user) ))) return NULL;
268 user->callback = func;
269 user->private = private;
271 /* Now insert it in the linked list */
273 LIST_FOR_EACH( ptr, &timeout_list )
275 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
276 if (!time_before( &timeout->when, when )) break;
278 list_add_before( ptr, &user->entry );
282 /* remove a timeout user */
283 void remove_timeout_user( struct timeout_user *user )
285 list_remove( &user->entry );
289 /* add a timeout in milliseconds to an absolute time */
290 void add_timeout( struct timeval *when, int timeout )
294 long sec = timeout / 1000;
295 if ((when->tv_usec += (timeout - 1000*sec) * 1000) >= 1000000)
297 when->tv_usec -= 1000000;
305 /****************************************************************/
308 static struct fd **poll_users; /* users array */
309 static struct pollfd *pollfd; /* poll fd array */
310 static int nb_users; /* count of array entries actually in use */
311 static int active_users; /* current number of active users */
312 static int allocated_users; /* count of allocated entries in the array */
313 static struct fd **freelist; /* list of free entries in the array */
318 static struct epoll_event *epoll_events;
320 /* set the events that epoll waits for on this fd; helper for set_fd_events */
321 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
323 struct epoll_event ev;
326 if (epoll_fd == -1) return;
328 if (events == -1) /* stop waiting on this fd completely */
330 if (pollfd[user].fd == -1) return; /* already removed */
333 else if (pollfd[user].fd == -1)
335 if (pollfd[user].events) return; /* stopped waiting on it, don't restart */
340 if (pollfd[user].events == events) return; /* nothing to do */
347 if (epoll_ctl( epoll_fd, ctl, fd->unix_fd, &ev ) == -1)
349 if (errno == ENOMEM) /* not enough memory, give up on epoll */
354 else perror( "epoll_ctl" ); /* should not happen */
358 #else /* USE_EPOLL */
360 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
364 #endif /* USE_EPOLL */
367 /* add a user in the poll array and return its index, or -1 on failure */
368 static int add_poll_user( struct fd *fd )
373 ret = freelist - poll_users;
374 freelist = (struct fd **)poll_users[ret];
378 if (nb_users == allocated_users)
380 struct fd **newusers;
381 struct pollfd *newpoll;
382 int new_count = allocated_users ? (allocated_users + allocated_users / 2) : 16;
383 if (!(newusers = realloc( poll_users, new_count * sizeof(*poll_users) ))) return -1;
384 if (!(newpoll = realloc( pollfd, new_count * sizeof(*pollfd) )))
387 poll_users = newusers;
392 poll_users = newusers;
395 if (!allocated_users) epoll_fd = epoll_create( new_count );
398 struct epoll_event *new_events;
399 if (!(new_events = realloc( epoll_events, new_count * sizeof(*epoll_events) )))
401 epoll_events = new_events;
404 allocated_users = new_count;
409 pollfd[ret].events = 0;
410 pollfd[ret].revents = 0;
411 poll_users[ret] = fd;
416 /* remove a user from the poll list */
417 static void remove_poll_user( struct fd *fd, int user )
420 assert( poll_users[user] == fd );
423 if (epoll_fd != -1 && pollfd[user].fd != -1)
425 struct epoll_event dummy;
426 epoll_ctl( epoll_fd, EPOLL_CTL_DEL, fd->unix_fd, &dummy );
429 pollfd[user].fd = -1;
430 pollfd[user].events = 0;
431 pollfd[user].revents = 0;
432 poll_users[user] = (struct fd *)freelist;
433 freelist = &poll_users[user];
437 /* process pending timeouts and return the time until the next timeout, in milliseconds */
438 static int get_next_timeout(void)
440 if (!list_empty( &timeout_list ))
442 struct list expired_list, *ptr;
445 gettimeofday( &now, NULL );
447 /* first remove all expired timers from the list */
449 list_init( &expired_list );
450 while ((ptr = list_head( &timeout_list )) != NULL)
452 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
454 if (!time_before( &now, &timeout->when ))
456 list_remove( &timeout->entry );
457 list_add_tail( &expired_list, &timeout->entry );
462 /* now call the callback for all the removed timers */
464 while ((ptr = list_head( &expired_list )) != NULL)
466 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
467 list_remove( &timeout->entry );
468 timeout->callback( timeout->private );
472 if ((ptr = list_head( &timeout_list )) != NULL)
474 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
475 int diff = (timeout->when.tv_sec - now.tv_sec) * 1000
476 + (timeout->when.tv_usec - now.tv_usec) / 1000;
477 if (diff < 0) diff = 0;
481 return -1; /* no pending timeouts */
484 /* server main poll() loop */
490 assert( POLLIN == EPOLLIN );
491 assert( POLLOUT == EPOLLOUT );
492 assert( POLLERR == EPOLLERR );
493 assert( POLLHUP == EPOLLHUP );
499 timeout = get_next_timeout();
501 if (!active_users) break; /* last user removed by a timeout */
502 if (epoll_fd == -1) break; /* an error occurred with epoll */
504 ret = epoll_wait( epoll_fd, epoll_events, allocated_users, timeout );
506 /* put the events into the pollfd array first, like poll does */
507 for (i = 0; i < ret; i++)
509 int user = epoll_events[i].data.u32;
510 pollfd[user].revents = epoll_events[i].events;
513 /* read events from the pollfd array, as set_fd_events may modify them */
514 for (i = 0; i < ret; i++)
516 int user = epoll_events[i].data.u32;
517 if (pollfd[user].revents) fd_poll_event( poll_users[user], pollfd[user].revents );
521 /* fall through to normal poll loop */
522 #endif /* USE_EPOLL */
526 timeout = get_next_timeout();
528 if (!active_users) break; /* last user removed by a timeout */
530 ret = poll( pollfd, nb_users, timeout );
533 for (i = 0; i < nb_users; i++)
535 if (pollfd[i].revents)
537 fd_poll_event( poll_users[i], pollfd[i].revents );
546 /****************************************************************/
547 /* inode functions */
551 static struct list inode_hash[HASH_SIZE];
553 /* close all pending file descriptors in the closed list */
554 static void inode_close_pending( struct inode *inode )
556 struct list *ptr = list_head( &inode->closed );
560 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
561 struct list *next = list_next( &inode->closed, ptr );
568 if (!fd->unlink) /* get rid of it unless there's an unlink pending on that file */
578 static void inode_dump( struct object *obj, int verbose )
580 struct inode *inode = (struct inode *)obj;
581 fprintf( stderr, "Inode dev=" );
582 DUMP_LONG_LONG( inode->dev );
583 fprintf( stderr, " ino=" );
584 DUMP_LONG_LONG( inode->ino );
585 fprintf( stderr, "\n" );
588 static void inode_destroy( struct object *obj )
590 struct inode *inode = (struct inode *)obj;
593 assert( list_empty(&inode->open) );
594 assert( list_empty(&inode->locks) );
596 list_remove( &inode->entry );
598 while ((ptr = list_head( &inode->closed )))
600 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
602 if (fd->fd != -1) close( fd->fd );
605 /* make sure it is still the same file */
607 if (!stat( fd->unlink, &st ) && st.st_dev == inode->dev && st.st_ino == inode->ino)
609 if (S_ISDIR(st.st_mode)) rmdir( fd->unlink );
610 else unlink( fd->unlink );
617 /* retrieve the inode object for a given fd, creating it if needed */
618 static struct inode *get_inode( dev_t dev, ino_t ino )
622 unsigned int hash = (dev ^ ino) % HASH_SIZE;
624 if (inode_hash[hash].next)
626 LIST_FOR_EACH( ptr, &inode_hash[hash] )
628 inode = LIST_ENTRY( ptr, struct inode, entry );
629 if (inode->dev == dev && inode->ino == ino)
630 return (struct inode *)grab_object( inode );
633 else list_init( &inode_hash[hash] );
635 /* not found, create it */
636 if ((inode = alloc_object( &inode_ops )))
641 list_init( &inode->open );
642 list_init( &inode->locks );
643 list_init( &inode->closed );
644 list_add_head( &inode_hash[hash], &inode->entry );
649 /* add fd to the indoe list of file descriptors to close */
650 static void inode_add_closed_fd( struct inode *inode, struct closed_fd *fd )
652 if (!list_empty( &inode->locks ))
654 list_add_head( &inode->closed, &fd->entry );
656 else if (fd->unlink[0]) /* close the fd but keep the structure around for unlink */
660 list_add_head( &inode->closed, &fd->entry );
662 else /* no locks on this inode and no unlink, get rid of the fd */
670 /****************************************************************/
671 /* file lock functions */
673 static void file_lock_dump( struct object *obj, int verbose )
675 struct file_lock *lock = (struct file_lock *)obj;
676 fprintf( stderr, "Lock %s fd=%p proc=%p start=",
677 lock->shared ? "shared" : "excl", lock->fd, lock->process );
678 DUMP_LONG_LONG( lock->start );
679 fprintf( stderr, " end=" );
680 DUMP_LONG_LONG( lock->end );
681 fprintf( stderr, "\n" );
684 static int file_lock_signaled( struct object *obj, struct thread *thread )
686 struct file_lock *lock = (struct file_lock *)obj;
687 /* lock is signaled if it has lost its owner */
688 return !lock->process;
691 /* set (or remove) a Unix lock if possible for the given range */
692 static int set_unix_lock( struct fd *fd, file_pos_t start, file_pos_t end, int type )
696 if (!fd->fs_locks) return 1; /* no fs locks possible for this fd */
699 if (start == end) return 1; /* can't set zero-byte lock */
700 if (start > max_unix_offset) return 1; /* ignore it */
702 fl.l_whence = SEEK_SET;
704 if (!end || end > max_unix_offset) fl.l_len = 0;
705 else fl.l_len = end - start;
706 if (fcntl( fd->unix_fd, F_SETLK, &fl ) != -1) return 1;
711 /* check whether locks work at all on this file system */
712 if (fcntl( fd->unix_fd, F_GETLK, &fl ) != -1)
714 set_error( STATUS_FILE_LOCK_CONFLICT );
720 /* no locking on this fs, just ignore it */
724 set_error( STATUS_FILE_LOCK_CONFLICT );
727 /* this can happen if we try to set a write lock on a read-only file */
728 /* we just ignore that error */
729 if (fl.l_type == F_WRLCK) return 1;
730 set_error( STATUS_ACCESS_DENIED );
736 /* this can happen if off_t is 64-bit but the kernel only supports 32-bit */
737 /* in that case we shrink the limit and retry */
738 if (max_unix_offset > INT_MAX)
740 max_unix_offset = INT_MAX;
751 /* check if interval [start;end) overlaps the lock */
752 inline static int lock_overlaps( struct file_lock *lock, file_pos_t start, file_pos_t end )
754 if (lock->end && start >= lock->end) return 0;
755 if (end && lock->start >= end) return 0;
759 /* remove Unix locks for all bytes in the specified area that are no longer locked */
760 static void remove_unix_locks( struct fd *fd, file_pos_t start, file_pos_t end )
768 } *first, *cur, *next, *buffer;
773 if (!fd->inode) return;
774 if (!fd->fs_locks) return;
775 if (start == end || start > max_unix_offset) return;
776 if (!end || end > max_unix_offset) end = max_unix_offset + 1;
778 /* count the number of locks overlapping the specified area */
780 LIST_FOR_EACH( ptr, &fd->inode->locks )
782 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
783 if (lock->start == lock->end) continue;
784 if (lock_overlaps( lock, start, end )) count++;
787 if (!count) /* no locks at all, we can unlock everything */
789 set_unix_lock( fd, start, end, F_UNLCK );
793 /* allocate space for the list of holes */
794 /* max. number of holes is number of locks + 1 */
796 if (!(buffer = malloc( sizeof(*buffer) * (count+1) ))) return;
800 first->start = start;
804 /* build a sorted list of unlocked holes in the specified area */
806 LIST_FOR_EACH( ptr, &fd->inode->locks )
808 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
809 if (lock->start == lock->end) continue;
810 if (!lock_overlaps( lock, start, end )) continue;
812 /* go through all the holes touched by this lock */
813 for (cur = first; cur; cur = cur->next)
815 if (cur->end <= lock->start) continue; /* hole is before start of lock */
816 if (lock->end && cur->start >= lock->end) break; /* hole is after end of lock */
818 /* now we know that lock is overlapping hole */
820 if (cur->start >= lock->start) /* lock starts before hole, shrink from start */
822 cur->start = lock->end;
823 if (cur->start && cur->start < cur->end) break; /* done with this lock */
824 /* now hole is empty, remove it */
825 if (cur->next) cur->next->prev = cur->prev;
826 if (cur->prev) cur->prev->next = cur->next;
827 else if (!(first = cur->next)) goto done; /* no more holes at all */
829 else if (!lock->end || cur->end <= lock->end) /* lock larger than hole, shrink from end */
831 cur->end = lock->start;
832 assert( cur->start < cur->end );
834 else /* lock is in the middle of hole, split hole in two */
837 next->next = cur->next;
839 next->start = lock->end;
840 next->end = cur->end;
841 cur->end = lock->start;
842 assert( next->start < next->end );
843 assert( cur->end < next->start );
845 break; /* done with this lock */
850 /* clear Unix locks for all the holes */
852 for (cur = first; cur; cur = cur->next)
853 set_unix_lock( fd, cur->start, cur->end, F_UNLCK );
859 /* create a new lock on a fd */
860 static struct file_lock *add_lock( struct fd *fd, int shared, file_pos_t start, file_pos_t end )
862 struct file_lock *lock;
864 if (!fd->inode) /* not a regular file */
866 set_error( STATUS_INVALID_HANDLE );
870 if (!(lock = alloc_object( &file_lock_ops ))) return NULL;
871 lock->shared = shared;
875 lock->process = current->process;
877 /* now try to set a Unix lock */
878 if (!set_unix_lock( lock->fd, lock->start, lock->end, lock->shared ? F_RDLCK : F_WRLCK ))
880 release_object( lock );
883 list_add_head( &fd->locks, &lock->fd_entry );
884 list_add_head( &fd->inode->locks, &lock->inode_entry );
885 list_add_head( &lock->process->locks, &lock->proc_entry );
889 /* remove an existing lock */
890 static void remove_lock( struct file_lock *lock, int remove_unix )
892 struct inode *inode = lock->fd->inode;
894 list_remove( &lock->fd_entry );
895 list_remove( &lock->inode_entry );
896 list_remove( &lock->proc_entry );
897 if (remove_unix) remove_unix_locks( lock->fd, lock->start, lock->end );
898 if (list_empty( &inode->locks )) inode_close_pending( inode );
899 lock->process = NULL;
900 wake_up( &lock->obj, 0 );
901 release_object( lock );
904 /* remove all locks owned by a given process */
905 void remove_process_locks( struct process *process )
909 while ((ptr = list_head( &process->locks )))
911 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, proc_entry );
912 remove_lock( lock, 1 ); /* this removes it from the list */
916 /* remove all locks on a given fd */
917 static void remove_fd_locks( struct fd *fd )
919 file_pos_t start = FILE_POS_T_MAX, end = 0;
922 while ((ptr = list_head( &fd->locks )))
924 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
925 if (lock->start < start) start = lock->start;
926 if (!lock->end || lock->end > end) end = lock->end - 1;
927 remove_lock( lock, 0 );
929 if (start < end) remove_unix_locks( fd, start, end + 1 );
932 /* add a lock on an fd */
933 /* returns handle to wait on */
934 obj_handle_t lock_fd( struct fd *fd, file_pos_t start, file_pos_t count, int shared, int wait )
937 file_pos_t end = start + count;
939 /* don't allow wrapping locks */
940 if (end && end < start)
942 set_error( STATUS_INVALID_PARAMETER );
946 /* check if another lock on that file overlaps the area */
947 LIST_FOR_EACH( ptr, &fd->inode->locks )
949 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
950 if (!lock_overlaps( lock, start, end )) continue;
951 if (lock->shared && shared) continue;
955 set_error( STATUS_FILE_LOCK_CONFLICT );
958 set_error( STATUS_PENDING );
959 return alloc_handle( current->process, lock, SYNCHRONIZE, 0 );
962 /* not found, add it */
963 if (add_lock( fd, shared, start, end )) return 0;
964 if (get_error() == STATUS_FILE_LOCK_CONFLICT)
966 /* Unix lock conflict -> tell client to wait and retry */
967 if (wait) set_error( STATUS_PENDING );
972 /* remove a lock on an fd */
973 void unlock_fd( struct fd *fd, file_pos_t start, file_pos_t count )
976 file_pos_t end = start + count;
978 /* find an existing lock with the exact same parameters */
979 LIST_FOR_EACH( ptr, &fd->locks )
981 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
982 if ((lock->start == start) && (lock->end == end))
984 remove_lock( lock, 1 );
988 set_error( STATUS_FILE_LOCK_CONFLICT );
992 /****************************************************************/
993 /* asynchronous operations support */
997 struct thread *thread;
1001 struct timeout_user *timeout;
1005 /* notifies client thread of new status of its async request */
1006 /* destroys the server side of it */
1007 static void async_terminate( struct async *async, int status )
1009 thread_queue_apc( async->thread, NULL, async->apc, APC_ASYNC_IO,
1010 1, async->user, async->sb, (void *)status );
1012 if (async->timeout) remove_timeout_user( async->timeout );
1013 async->timeout = NULL;
1014 list_remove( &async->entry );
1015 release_object( async->thread );
1019 /* cb for timeout on an async request */
1020 static void async_callback(void *private)
1022 struct async *async = (struct async *)private;
1024 /* fprintf(stderr, "async timeout out %p\n", async); */
1025 async->timeout = NULL;
1026 async_terminate( async, STATUS_TIMEOUT );
1029 /* create an async on a given queue of a fd */
1030 struct async *create_async(struct thread *thread, int* timeout, struct list *queue,
1031 void *io_apc, void *io_user, void* io_sb)
1033 struct async *async = mem_alloc( sizeof(struct async) );
1035 if (!async) return NULL;
1037 async->thread = (struct thread *)grab_object(thread);
1038 async->apc = io_apc;
1039 async->user = io_user;
1042 list_add_tail( queue, &async->entry );
1046 struct timeval when;
1048 gettimeofday( &when, NULL );
1049 add_timeout( &when, *timeout );
1050 async->timeout = add_timeout_user( &when, async_callback, async );
1052 else async->timeout = NULL;
1057 /* terminate the async operation at the head of the queue */
1058 void async_terminate_head( struct list *queue, int status )
1060 struct list *ptr = list_head( queue );
1061 if (ptr) async_terminate( LIST_ENTRY( ptr, struct async, entry ), status );
1064 /****************************************************************/
1065 /* file descriptor functions */
1067 static void fd_dump( struct object *obj, int verbose )
1069 struct fd *fd = (struct fd *)obj;
1070 fprintf( stderr, "Fd unix_fd=%d user=%p", fd->unix_fd, fd->user );
1071 if (fd->inode) fprintf( stderr, " inode=%p unlink='%s'", fd->inode, fd->closed->unlink );
1072 fprintf( stderr, "\n" );
1075 static void fd_destroy( struct object *obj )
1077 struct fd *fd = (struct fd *)obj;
1079 async_terminate_queue( &fd->read_q, STATUS_CANCELLED );
1080 async_terminate_queue( &fd->write_q, STATUS_CANCELLED );
1082 remove_fd_locks( fd );
1083 list_remove( &fd->inode_entry );
1084 if (fd->poll_index != -1) remove_poll_user( fd, fd->poll_index );
1087 inode_add_closed_fd( fd->inode, fd->closed );
1088 release_object( fd->inode );
1090 else /* no inode, close it right away */
1092 if (fd->unix_fd != -1) close( fd->unix_fd );
1096 /* set the events that select waits for on this fd */
1097 void set_fd_events( struct fd *fd, int events )
1099 int user = fd->poll_index;
1100 assert( poll_users[user] == fd );
1102 set_fd_epoll_events( fd, user, events );
1104 if (events == -1) /* stop waiting on this fd completely */
1106 pollfd[user].fd = -1;
1107 pollfd[user].events = POLLERR;
1108 pollfd[user].revents = 0;
1110 else if (pollfd[user].fd != -1 || !pollfd[user].events)
1112 pollfd[user].fd = fd->unix_fd;
1113 pollfd[user].events = events;
1117 /* allocate an fd object, without setting the unix fd yet */
1118 struct fd *alloc_fd( const struct fd_ops *fd_user_ops, struct object *user )
1120 struct fd *fd = alloc_object( &fd_ops );
1122 if (!fd) return NULL;
1124 fd->fd_ops = fd_user_ops;
1132 fd->poll_index = -1;
1133 list_init( &fd->inode_entry );
1134 list_init( &fd->locks );
1135 list_init( &fd->read_q );
1136 list_init( &fd->write_q );
1138 if ((fd->poll_index = add_poll_user( fd )) == -1)
1140 release_object( fd );
1146 /* check if the desired access is possible without violating */
1147 /* the sharing mode of other opens of the same file */
1148 static int check_sharing( struct fd *fd, unsigned int access, unsigned int sharing )
1150 unsigned int existing_sharing = FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
1151 unsigned int existing_access = 0;
1155 /* if access mode is 0, sharing mode is ignored */
1156 if (!access) sharing = existing_sharing;
1157 fd->access = access;
1158 fd->sharing = sharing;
1160 LIST_FOR_EACH( ptr, &fd->inode->open )
1162 struct fd *fd_ptr = LIST_ENTRY( ptr, struct fd, inode_entry );
1165 existing_sharing &= fd_ptr->sharing;
1166 existing_access |= fd_ptr->access;
1167 if (fd_ptr->closed->unlink[0]) unlink = 1;
1171 if ((access & GENERIC_READ) && !(existing_sharing & FILE_SHARE_READ)) return 0;
1172 if ((access & GENERIC_WRITE) && !(existing_sharing & FILE_SHARE_WRITE)) return 0;
1173 if ((existing_access & GENERIC_READ) && !(sharing & FILE_SHARE_READ)) return 0;
1174 if ((existing_access & GENERIC_WRITE) && !(sharing & FILE_SHARE_WRITE)) return 0;
1175 if (fd->closed->unlink[0] && !(existing_sharing & FILE_SHARE_DELETE)) return 0;
1176 if (unlink && !(sharing & FILE_SHARE_DELETE)) return 0;
1180 /* open() wrapper using a struct fd */
1181 /* the fd must have been created with alloc_fd */
1182 /* on error the fd object is released */
1183 struct fd *open_fd( struct fd *fd, const char *name, int flags, mode_t *mode,
1184 unsigned int access, unsigned int sharing, unsigned int options )
1187 struct closed_fd *closed_fd;
1188 const char *unlink_name = "";
1190 assert( fd->unix_fd == -1 );
1192 if (options & FILE_DELETE_ON_CLOSE) unlink_name = name;
1193 if (!(closed_fd = mem_alloc( sizeof(*closed_fd) + strlen(unlink_name) )))
1195 release_object( fd );
1198 /* create the directory if needed */
1199 if ((options & FILE_DIRECTORY_FILE) && (flags & O_CREAT))
1201 if (mkdir( name, 0777 ) == -1)
1203 if (errno != EEXIST || (flags & O_EXCL))
1206 release_object( fd );
1211 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
1213 if ((fd->unix_fd = open( name, flags & ~O_TRUNC, *mode )) == -1)
1216 release_object( fd );
1220 closed_fd->fd = fd->unix_fd;
1221 closed_fd->unlink[0] = 0;
1222 fstat( fd->unix_fd, &st );
1225 /* only bother with an inode for normal files and directories */
1226 if (S_ISREG(st.st_mode) || S_ISDIR(st.st_mode))
1228 struct inode *inode = get_inode( st.st_dev, st.st_ino );
1232 /* we can close the fd because there are no others open on the same file,
1233 * otherwise we wouldn't have failed to allocate a new inode
1238 fd->closed = closed_fd;
1239 list_add_head( &inode->open, &fd->inode_entry );
1241 /* check directory options */
1242 if ((options & FILE_DIRECTORY_FILE) && !S_ISDIR(st.st_mode))
1244 release_object( fd );
1245 set_error( STATUS_NOT_A_DIRECTORY );
1248 if ((options & FILE_NON_DIRECTORY_FILE) && S_ISDIR(st.st_mode))
1250 release_object( fd );
1251 set_error( STATUS_FILE_IS_A_DIRECTORY );
1254 if (!check_sharing( fd, access, sharing ))
1256 release_object( fd );
1257 set_error( STATUS_SHARING_VIOLATION );
1260 strcpy( closed_fd->unlink, unlink_name );
1261 if (flags & O_TRUNC) ftruncate( fd->unix_fd, 0 );
1263 else /* special file */
1265 if (options & FILE_DIRECTORY_FILE)
1267 set_error( STATUS_NOT_A_DIRECTORY );
1270 if (unlink_name[0]) /* we can't unlink special files */
1272 set_error( STATUS_INVALID_PARAMETER );
1280 release_object( fd );
1285 /* create an fd for an anonymous file */
1286 /* if the function fails the unix fd is closed */
1287 struct fd *create_anonymous_fd( const struct fd_ops *fd_user_ops, int unix_fd, struct object *user )
1289 struct fd *fd = alloc_fd( fd_user_ops, user );
1293 fd->unix_fd = unix_fd;
1300 /* retrieve the object that is using an fd */
1301 void *get_fd_user( struct fd *fd )
1306 /* retrieve the unix fd for an object */
1307 int get_unix_fd( struct fd *fd )
1312 /* check if two file descriptors point to the same file */
1313 int is_same_file_fd( struct fd *fd1, struct fd *fd2 )
1315 return fd1->inode == fd2->inode;
1318 /* callback for event happening in the main poll() loop */
1319 void fd_poll_event( struct fd *fd, int event )
1321 return fd->fd_ops->poll_event( fd, event );
1324 /* check if events are pending and if yes return which one(s) */
1325 int check_fd_events( struct fd *fd, int events )
1329 pfd.fd = fd->unix_fd;
1330 pfd.events = events;
1331 if (poll( &pfd, 1, 0 ) <= 0) return 0;
1335 /* default add_queue() routine for objects that poll() on an fd */
1336 int default_fd_add_queue( struct object *obj, struct wait_queue_entry *entry )
1338 struct fd *fd = get_obj_fd( obj );
1341 if (list_empty( &obj->wait_queue )) /* first on the queue */
1342 set_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1343 add_queue( obj, entry );
1344 release_object( fd );
1348 /* default remove_queue() routine for objects that poll() on an fd */
1349 void default_fd_remove_queue( struct object *obj, struct wait_queue_entry *entry )
1351 struct fd *fd = get_obj_fd( obj );
1354 remove_queue( obj, entry );
1355 if (list_empty( &obj->wait_queue )) /* last on the queue is gone */
1356 set_fd_events( fd, 0 );
1357 release_object( obj );
1358 release_object( fd );
1361 /* default signaled() routine for objects that poll() on an fd */
1362 int default_fd_signaled( struct object *obj, struct thread *thread )
1365 struct fd *fd = get_obj_fd( obj );
1367 if (fd->inode) return 1; /* regular files are always signaled */
1369 events = fd->fd_ops->get_poll_events( fd );
1370 ret = check_fd_events( fd, events ) != 0;
1373 set_fd_events( fd, 0 ); /* stop waiting on select() if we are signaled */
1374 else if (!list_empty( &obj->wait_queue ))
1375 set_fd_events( fd, events ); /* restart waiting on poll() if we are no longer signaled */
1377 release_object( fd );
1381 int default_fd_get_poll_events( struct fd *fd )
1385 if (!list_empty( &fd->read_q ))
1387 if (!list_empty( &fd->write_q ))
1393 /* default handler for poll() events */
1394 void default_poll_event( struct fd *fd, int event )
1396 if (!list_empty( &fd->read_q ) && (POLLIN & event) )
1398 async_terminate_head( &fd->read_q, STATUS_ALERTED );
1401 if (!list_empty( &fd->write_q ) && (POLLOUT & event) )
1403 async_terminate_head( &fd->write_q, STATUS_ALERTED );
1407 /* if an error occurred, stop polling this fd to avoid busy-looping */
1408 if (event & (POLLERR | POLLHUP)) set_fd_events( fd, -1 );
1409 wake_up( fd->user, 0 );
1412 void default_fd_queue_async( struct fd *fd, void *apc, void *user, void *io_sb, int type, int count )
1417 if (!(fd->fd_ops->get_file_info( fd ) & FD_FLAG_OVERLAPPED))
1419 set_error( STATUS_INVALID_HANDLE );
1425 case ASYNC_TYPE_READ:
1426 queue = &fd->read_q;
1428 case ASYNC_TYPE_WRITE:
1429 queue = &fd->write_q;
1432 set_error( STATUS_INVALID_PARAMETER );
1436 if (!create_async( current, NULL, queue, apc, user, io_sb ))
1439 /* Check if the new pending request can be served immediately */
1440 events = check_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1441 if (events) fd->fd_ops->poll_event( fd, events );
1443 set_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1446 void default_fd_cancel_async( struct fd *fd )
1448 async_terminate_queue( &fd->read_q, STATUS_CANCELLED );
1449 async_terminate_queue( &fd->write_q, STATUS_CANCELLED );
1452 /* default flush() routine */
1453 int no_flush( struct fd *fd, struct event **event )
1455 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1459 /* default get_file_info() routine */
1460 int no_get_file_info( struct fd *fd )
1462 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1466 /* default queue_async() routine */
1467 void no_queue_async( struct fd *fd, void* apc, void* user, void* io_sb,
1468 int type, int count)
1470 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1473 /* default cancel_async() routine */
1474 void no_cancel_async( struct fd *fd )
1476 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1479 /* same as get_handle_obj but retrieve the struct fd associated to the object */
1480 static struct fd *get_handle_fd_obj( struct process *process, obj_handle_t handle,
1481 unsigned int access )
1483 struct fd *fd = NULL;
1486 if ((obj = get_handle_obj( process, handle, access, NULL )))
1488 fd = get_obj_fd( obj );
1489 release_object( obj );
1494 /* flush a file buffers */
1495 DECL_HANDLER(flush_file)
1497 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1498 struct event * event = NULL;
1502 fd->fd_ops->flush( fd, &event );
1505 reply->event = alloc_handle( current->process, event, SYNCHRONIZE, 0 );
1507 release_object( fd );
1511 /* get a Unix fd to access a file */
1512 DECL_HANDLER(get_handle_fd)
1518 if ((fd = get_handle_fd_obj( current->process, req->handle, req->access )))
1520 int unix_fd = get_handle_unix_fd( current->process, req->handle, req->access );
1521 if (unix_fd != -1) reply->fd = unix_fd;
1522 else if (!get_error())
1524 assert( fd->unix_fd != -1 );
1525 send_client_fd( current->process, fd->unix_fd, req->handle );
1527 reply->flags = fd->fd_ops->get_file_info( fd );
1528 release_object( fd );
1532 /* create / reschedule an async I/O */
1533 DECL_HANDLER(register_async)
1535 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1538 * The queue_async method must do the following:
1540 * 1. Get the async_queue for the request of given type.
1541 * 2. Create a new asynchronous request for the selected queue
1542 * 3. Carry out any operations necessary to adjust the object's poll events
1543 * Usually: set_elect_events (obj, obj->ops->get_poll_events()).
1544 * 4. When the async request is triggered, then send back (with a proper APC)
1545 * the trigger (STATUS_ALERTED) to the thread that posted the request.
1546 * async_destroy() is to be called: it will both notify the sender about
1547 * the trigger and destroy the request by itself
1548 * See also the implementations in file.c, serial.c, and sock.c.
1553 fd->fd_ops->queue_async( fd, req->io_apc, req->io_user, req->io_sb,
1554 req->type, req->count );
1555 release_object( fd );
1559 /* cancels all async I/O */
1560 DECL_HANDLER(cancel_async)
1562 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1565 /* Note: we don't kill the queued APC_ASYNC_IO on this thread because
1566 * NtCancelIoFile() will force the pending APC to be run. Since,
1567 * Windows only guarantees that the current thread will have no async
1568 * operation on the current fd when NtCancelIoFile returns, this shall
1571 fd->fd_ops->cancel_async( fd );
1572 release_object( fd );