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>
57 #if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
58 # include <sys/epoll.h>
60 #elif defined(linux) && defined(__i386__) && defined(HAVE_STDINT_H)
62 # define EPOLLIN POLLIN
63 # define EPOLLOUT POLLOUT
64 # define EPOLLERR POLLERR
65 # define EPOLLHUP POLLHUP
66 # define EPOLL_CTL_ADD 1
67 # define EPOLL_CTL_DEL 2
68 # define EPOLL_CTL_MOD 3
70 typedef union epoll_data
84 #define SYSCALL_RET(ret) do { \
85 if (ret < 0) { errno = -ret; ret = -1; } \
89 static inline int epoll_create( int size )
92 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
93 : "=a" (ret) : "0" (254 /*NR_epoll_create*/), "r" (size) );
97 static inline int epoll_ctl( int epfd, int op, int fd, const struct epoll_event *event )
100 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
102 : "0" (255 /*NR_epoll_ctl*/), "r" (epfd), "c" (op), "d" (fd), "S" (event), "m" (*event) );
106 static inline int epoll_wait( int epfd, struct epoll_event *events, int maxevents, int timeout )
109 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
111 : "0" (256 /*NR_epoll_wait*/), "r" (epfd), "c" (events), "d" (maxevents), "S" (timeout)
117 #endif /* linux && __i386__ && HAVE_STDINT_H */
120 /* Because of the stupid Posix locking semantics, we need to keep
121 * track of all file descriptors referencing a given file, and not
122 * close a single one until all the locks are gone (sigh).
125 /* file descriptor object */
127 /* closed_fd is used to keep track of the unix fd belonging to a closed fd object */
130 struct list entry; /* entry in inode closed list */
131 int fd; /* the unix file descriptor */
132 char unlink[1]; /* name to unlink on close (if any) */
137 struct object obj; /* object header */
138 const struct fd_ops *fd_ops; /* file descriptor operations */
139 struct inode *inode; /* inode that this fd belongs to */
140 struct list inode_entry; /* entry in inode fd list */
141 struct closed_fd *closed; /* structure to store the unix fd at destroy time */
142 struct object *user; /* object using this file descriptor */
143 struct list locks; /* list of locks on this fd */
144 unsigned int access; /* file access (GENERIC_READ/WRITE) */
145 unsigned int sharing; /* file sharing mode */
146 int unix_fd; /* unix file descriptor */
147 int fs_locks; /* can we use filesystem locks for this fd? */
148 int poll_index; /* index of fd in poll array */
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 fd_destroy /* destroy */
171 struct object obj; /* object header */
172 struct list entry; /* inode hash list entry */
173 unsigned int hash; /* hashing code */
174 dev_t dev; /* device number */
175 ino_t ino; /* inode number */
176 struct list open; /* list of open file descriptors */
177 struct list locks; /* list of file locks */
178 struct list closed; /* list of file descriptors to close at destroy time */
181 static void inode_dump( struct object *obj, int verbose );
182 static void inode_destroy( struct object *obj );
184 static const struct object_ops inode_ops =
186 sizeof(struct inode), /* size */
187 inode_dump, /* dump */
188 no_add_queue, /* add_queue */
189 NULL, /* remove_queue */
191 NULL, /* satisfied */
192 no_signal, /* signal */
193 no_get_fd, /* get_fd */
194 inode_destroy /* destroy */
197 /* file lock object */
201 struct object obj; /* object header */
202 struct fd *fd; /* fd owning this lock */
203 struct list fd_entry; /* entry in list of locks on a given fd */
204 struct list inode_entry; /* entry in inode list of locks */
205 int shared; /* shared lock? */
206 file_pos_t start; /* locked region is interval [start;end) */
208 struct process *process; /* process owning this lock */
209 struct list proc_entry; /* entry in list of locks owned by the process */
212 static void file_lock_dump( struct object *obj, int verbose );
213 static int file_lock_signaled( struct object *obj, struct thread *thread );
215 static const struct object_ops file_lock_ops =
217 sizeof(struct file_lock), /* size */
218 file_lock_dump, /* dump */
219 add_queue, /* add_queue */
220 remove_queue, /* remove_queue */
221 file_lock_signaled, /* signaled */
222 no_satisfied, /* satisfied */
223 no_signal, /* signal */
224 no_get_fd, /* get_fd */
225 no_destroy /* destroy */
229 #define OFF_T_MAX (~((file_pos_t)1 << (8*sizeof(off_t)-1)))
230 #define FILE_POS_T_MAX (~(file_pos_t)0)
232 static file_pos_t max_unix_offset = OFF_T_MAX;
234 #define DUMP_LONG_LONG(val) do { \
235 if (sizeof(val) > sizeof(unsigned long) && (val) > ~0UL) \
236 fprintf( stderr, "%lx%08lx", (unsigned long)((val) >> 32), (unsigned long)(val) ); \
238 fprintf( stderr, "%lx", (unsigned long)(val) ); \
243 /****************************************************************/
244 /* timeouts support */
248 struct list entry; /* entry in sorted timeout list */
249 struct timeval when; /* timeout expiry (absolute time) */
250 timeout_callback callback; /* callback function */
251 void *private; /* callback private data */
254 static struct list timeout_list = LIST_INIT(timeout_list); /* sorted timeouts list */
256 /* add a timeout user */
257 struct timeout_user *add_timeout_user( const struct timeval *when, timeout_callback func,
260 struct timeout_user *user;
263 if (!(user = mem_alloc( sizeof(*user) ))) return NULL;
265 user->callback = func;
266 user->private = private;
268 /* Now insert it in the linked list */
270 LIST_FOR_EACH( ptr, &timeout_list )
272 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
273 if (!time_before( &timeout->when, when )) break;
275 list_add_before( ptr, &user->entry );
279 /* remove a timeout user */
280 void remove_timeout_user( struct timeout_user *user )
282 list_remove( &user->entry );
286 /* add a timeout in milliseconds to an absolute time */
287 void add_timeout( struct timeval *when, int timeout )
291 long sec = timeout / 1000;
292 if ((when->tv_usec += (timeout - 1000*sec) * 1000) >= 1000000)
294 when->tv_usec -= 1000000;
302 /****************************************************************/
305 static struct fd **poll_users; /* users array */
306 static struct pollfd *pollfd; /* poll fd array */
307 static int nb_users; /* count of array entries actually in use */
308 static int active_users; /* current number of active users */
309 static int allocated_users; /* count of allocated entries in the array */
310 static struct fd **freelist; /* list of free entries in the array */
315 static struct epoll_event *epoll_events;
317 /* set the events that epoll waits for on this fd; helper for set_fd_events */
318 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
320 struct epoll_event ev;
323 if (epoll_fd == -1) return;
325 if (events == -1) /* stop waiting on this fd completely */
327 if (pollfd[user].fd == -1) return; /* already removed */
330 else if (pollfd[user].fd == -1)
332 if (pollfd[user].events) return; /* stopped waiting on it, don't restart */
337 if (pollfd[user].events == events) return; /* nothing to do */
344 if (epoll_ctl( epoll_fd, ctl, fd->unix_fd, &ev ) == -1)
346 if (errno == ENOMEM) /* not enough memory, give up on epoll */
351 else perror( "epoll_ctl" ); /* should not happen */
355 #else /* USE_EPOLL */
357 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
361 #endif /* USE_EPOLL */
364 /* add a user in the poll array and return its index, or -1 on failure */
365 static int add_poll_user( struct fd *fd )
370 ret = freelist - poll_users;
371 freelist = (struct fd **)poll_users[ret];
375 if (nb_users == allocated_users)
377 struct fd **newusers;
378 struct pollfd *newpoll;
379 int new_count = allocated_users ? (allocated_users + allocated_users / 2) : 16;
380 if (!(newusers = realloc( poll_users, new_count * sizeof(*poll_users) ))) return -1;
381 if (!(newpoll = realloc( pollfd, new_count * sizeof(*pollfd) )))
384 poll_users = newusers;
389 poll_users = newusers;
392 if (!allocated_users) epoll_fd = epoll_create( new_count );
395 struct epoll_event *new_events;
396 if (!(new_events = realloc( epoll_events, new_count * sizeof(*epoll_events) )))
398 epoll_events = new_events;
401 allocated_users = new_count;
406 pollfd[ret].events = 0;
407 pollfd[ret].revents = 0;
408 poll_users[ret] = fd;
413 /* remove a user from the poll list */
414 static void remove_poll_user( struct fd *fd, int user )
417 assert( poll_users[user] == fd );
420 if (epoll_fd != -1 && pollfd[user].fd != -1)
422 struct epoll_event dummy;
423 epoll_ctl( epoll_fd, EPOLL_CTL_DEL, fd->unix_fd, &dummy );
426 pollfd[user].fd = -1;
427 pollfd[user].events = 0;
428 pollfd[user].revents = 0;
429 poll_users[user] = (struct fd *)freelist;
430 freelist = &poll_users[user];
434 /* process pending timeouts and return the time until the next timeout, in milliseconds */
435 static int get_next_timeout(void)
437 if (!list_empty( &timeout_list ))
439 struct list expired_list, *ptr;
442 gettimeofday( &now, NULL );
444 /* first remove all expired timers from the list */
446 list_init( &expired_list );
447 while ((ptr = list_head( &timeout_list )) != NULL)
449 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
451 if (!time_before( &now, &timeout->when ))
453 list_remove( &timeout->entry );
454 list_add_tail( &expired_list, &timeout->entry );
459 /* now call the callback for all the removed timers */
461 while ((ptr = list_head( &expired_list )) != NULL)
463 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
464 list_remove( &timeout->entry );
465 timeout->callback( timeout->private );
469 if ((ptr = list_head( &timeout_list )) != NULL)
471 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
472 int diff = (timeout->when.tv_sec - now.tv_sec) * 1000
473 + (timeout->when.tv_usec - now.tv_usec) / 1000;
474 if (diff < 0) diff = 0;
478 return -1; /* no pending timeouts */
481 /* server main poll() loop */
487 assert( POLLIN == EPOLLIN );
488 assert( POLLOUT == EPOLLOUT );
489 assert( POLLERR == EPOLLERR );
490 assert( POLLHUP == EPOLLHUP );
496 timeout = get_next_timeout();
498 if (!active_users) break; /* last user removed by a timeout */
499 if (epoll_fd == -1) break; /* an error occurred with epoll */
501 ret = epoll_wait( epoll_fd, epoll_events, allocated_users, timeout );
503 /* put the events into the pollfd array first, like poll does */
504 for (i = 0; i < ret; i++)
506 int user = epoll_events[i].data.u32;
507 pollfd[user].revents = epoll_events[i].events;
510 /* read events from the pollfd array, as set_fd_events may modify them */
511 for (i = 0; i < ret; i++)
513 int user = epoll_events[i].data.u32;
514 if (pollfd[user].revents) fd_poll_event( poll_users[user], pollfd[user].revents );
518 /* fall through to normal poll loop */
519 #endif /* USE_EPOLL */
523 timeout = get_next_timeout();
525 if (!active_users) break; /* last user removed by a timeout */
527 ret = poll( pollfd, nb_users, timeout );
530 for (i = 0; i < nb_users; i++)
532 if (pollfd[i].revents)
534 fd_poll_event( poll_users[i], pollfd[i].revents );
543 /****************************************************************/
544 /* inode functions */
548 static struct list inode_hash[HASH_SIZE];
550 /* close all pending file descriptors in the closed list */
551 static void inode_close_pending( struct inode *inode )
553 struct list *ptr = list_head( &inode->closed );
557 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
558 struct list *next = list_next( &inode->closed, ptr );
565 if (!fd->unlink) /* get rid of it unless there's an unlink pending on that file */
575 static void inode_dump( struct object *obj, int verbose )
577 struct inode *inode = (struct inode *)obj;
578 fprintf( stderr, "Inode dev=" );
579 DUMP_LONG_LONG( inode->dev );
580 fprintf( stderr, " ino=" );
581 DUMP_LONG_LONG( inode->ino );
582 fprintf( stderr, "\n" );
585 static void inode_destroy( struct object *obj )
587 struct inode *inode = (struct inode *)obj;
590 assert( list_empty(&inode->open) );
591 assert( list_empty(&inode->locks) );
593 list_remove( &inode->entry );
595 while ((ptr = list_head( &inode->closed )))
597 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
599 if (fd->fd != -1) close( fd->fd );
602 /* make sure it is still the same file */
604 if (!stat( fd->unlink, &st ) && st.st_dev == inode->dev && st.st_ino == inode->ino)
606 if (S_ISDIR(st.st_mode)) rmdir( fd->unlink );
607 else unlink( fd->unlink );
614 /* retrieve the inode object for a given fd, creating it if needed */
615 static struct inode *get_inode( dev_t dev, ino_t ino )
619 unsigned int hash = (dev ^ ino) % HASH_SIZE;
621 if (inode_hash[hash].next)
623 LIST_FOR_EACH( ptr, &inode_hash[hash] )
625 inode = LIST_ENTRY( ptr, struct inode, entry );
626 if (inode->dev == dev && inode->ino == ino)
627 return (struct inode *)grab_object( inode );
630 else list_init( &inode_hash[hash] );
632 /* not found, create it */
633 if ((inode = alloc_object( &inode_ops )))
638 list_init( &inode->open );
639 list_init( &inode->locks );
640 list_init( &inode->closed );
641 list_add_head( &inode_hash[hash], &inode->entry );
646 /* add fd to the indoe list of file descriptors to close */
647 static void inode_add_closed_fd( struct inode *inode, struct closed_fd *fd )
649 if (!list_empty( &inode->locks ))
651 list_add_head( &inode->closed, &fd->entry );
653 else if (fd->unlink[0]) /* close the fd but keep the structure around for unlink */
657 list_add_head( &inode->closed, &fd->entry );
659 else /* no locks on this inode and no unlink, get rid of the fd */
667 /****************************************************************/
668 /* file lock functions */
670 static void file_lock_dump( struct object *obj, int verbose )
672 struct file_lock *lock = (struct file_lock *)obj;
673 fprintf( stderr, "Lock %s fd=%p proc=%p start=",
674 lock->shared ? "shared" : "excl", lock->fd, lock->process );
675 DUMP_LONG_LONG( lock->start );
676 fprintf( stderr, " end=" );
677 DUMP_LONG_LONG( lock->end );
678 fprintf( stderr, "\n" );
681 static int file_lock_signaled( struct object *obj, struct thread *thread )
683 struct file_lock *lock = (struct file_lock *)obj;
684 /* lock is signaled if it has lost its owner */
685 return !lock->process;
688 /* set (or remove) a Unix lock if possible for the given range */
689 static int set_unix_lock( struct fd *fd, file_pos_t start, file_pos_t end, int type )
693 if (!fd->fs_locks) return 1; /* no fs locks possible for this fd */
696 if (start == end) return 1; /* can't set zero-byte lock */
697 if (start > max_unix_offset) return 1; /* ignore it */
699 fl.l_whence = SEEK_SET;
701 if (!end || end > max_unix_offset) fl.l_len = 0;
702 else fl.l_len = end - start;
703 if (fcntl( fd->unix_fd, F_SETLK, &fl ) != -1) return 1;
708 /* check whether locks work at all on this file system */
709 if (fcntl( fd->unix_fd, F_GETLK, &fl ) != -1)
711 set_error( STATUS_FILE_LOCK_CONFLICT );
717 /* no locking on this fs, just ignore it */
721 set_error( STATUS_FILE_LOCK_CONFLICT );
724 /* this can happen if we try to set a write lock on a read-only file */
725 /* we just ignore that error */
726 if (fl.l_type == F_WRLCK) return 1;
727 set_error( STATUS_ACCESS_DENIED );
733 /* this can happen if off_t is 64-bit but the kernel only supports 32-bit */
734 /* in that case we shrink the limit and retry */
735 if (max_unix_offset > INT_MAX)
737 max_unix_offset = INT_MAX;
748 /* check if interval [start;end) overlaps the lock */
749 inline static int lock_overlaps( struct file_lock *lock, file_pos_t start, file_pos_t end )
751 if (lock->end && start >= lock->end) return 0;
752 if (end && lock->start >= end) return 0;
756 /* remove Unix locks for all bytes in the specified area that are no longer locked */
757 static void remove_unix_locks( struct fd *fd, file_pos_t start, file_pos_t end )
765 } *first, *cur, *next, *buffer;
770 if (!fd->inode) return;
771 if (!fd->fs_locks) return;
772 if (start == end || start > max_unix_offset) return;
773 if (!end || end > max_unix_offset) end = max_unix_offset + 1;
775 /* count the number of locks overlapping the specified area */
777 LIST_FOR_EACH( ptr, &fd->inode->locks )
779 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
780 if (lock->start == lock->end) continue;
781 if (lock_overlaps( lock, start, end )) count++;
784 if (!count) /* no locks at all, we can unlock everything */
786 set_unix_lock( fd, start, end, F_UNLCK );
790 /* allocate space for the list of holes */
791 /* max. number of holes is number of locks + 1 */
793 if (!(buffer = malloc( sizeof(*buffer) * (count+1) ))) return;
797 first->start = start;
801 /* build a sorted list of unlocked holes in the specified area */
803 LIST_FOR_EACH( ptr, &fd->inode->locks )
805 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
806 if (lock->start == lock->end) continue;
807 if (!lock_overlaps( lock, start, end )) continue;
809 /* go through all the holes touched by this lock */
810 for (cur = first; cur; cur = cur->next)
812 if (cur->end <= lock->start) continue; /* hole is before start of lock */
813 if (lock->end && cur->start >= lock->end) break; /* hole is after end of lock */
815 /* now we know that lock is overlapping hole */
817 if (cur->start >= lock->start) /* lock starts before hole, shrink from start */
819 cur->start = lock->end;
820 if (cur->start && cur->start < cur->end) break; /* done with this lock */
821 /* now hole is empty, remove it */
822 if (cur->next) cur->next->prev = cur->prev;
823 if (cur->prev) cur->prev->next = cur->next;
824 else if (!(first = cur->next)) goto done; /* no more holes at all */
826 else if (!lock->end || cur->end <= lock->end) /* lock larger than hole, shrink from end */
828 cur->end = lock->start;
829 assert( cur->start < cur->end );
831 else /* lock is in the middle of hole, split hole in two */
834 next->next = cur->next;
836 next->start = lock->end;
837 next->end = cur->end;
838 cur->end = lock->start;
839 assert( next->start < next->end );
840 assert( cur->end < next->start );
842 break; /* done with this lock */
847 /* clear Unix locks for all the holes */
849 for (cur = first; cur; cur = cur->next)
850 set_unix_lock( fd, cur->start, cur->end, F_UNLCK );
856 /* create a new lock on a fd */
857 static struct file_lock *add_lock( struct fd *fd, int shared, file_pos_t start, file_pos_t end )
859 struct file_lock *lock;
861 if (!fd->inode) /* not a regular file */
863 set_error( STATUS_INVALID_HANDLE );
867 if (!(lock = alloc_object( &file_lock_ops ))) return NULL;
868 lock->shared = shared;
872 lock->process = current->process;
874 /* now try to set a Unix lock */
875 if (!set_unix_lock( lock->fd, lock->start, lock->end, lock->shared ? F_RDLCK : F_WRLCK ))
877 release_object( lock );
880 list_add_head( &fd->locks, &lock->fd_entry );
881 list_add_head( &fd->inode->locks, &lock->inode_entry );
882 list_add_head( &lock->process->locks, &lock->proc_entry );
886 /* remove an existing lock */
887 static void remove_lock( struct file_lock *lock, int remove_unix )
889 struct inode *inode = lock->fd->inode;
891 list_remove( &lock->fd_entry );
892 list_remove( &lock->inode_entry );
893 list_remove( &lock->proc_entry );
894 if (remove_unix) remove_unix_locks( lock->fd, lock->start, lock->end );
895 if (list_empty( &inode->locks )) inode_close_pending( inode );
896 lock->process = NULL;
897 wake_up( &lock->obj, 0 );
898 release_object( lock );
901 /* remove all locks owned by a given process */
902 void remove_process_locks( struct process *process )
906 while ((ptr = list_head( &process->locks )))
908 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, proc_entry );
909 remove_lock( lock, 1 ); /* this removes it from the list */
913 /* remove all locks on a given fd */
914 static void remove_fd_locks( struct fd *fd )
916 file_pos_t start = FILE_POS_T_MAX, end = 0;
919 while ((ptr = list_head( &fd->locks )))
921 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
922 if (lock->start < start) start = lock->start;
923 if (!lock->end || lock->end > end) end = lock->end - 1;
924 remove_lock( lock, 0 );
926 if (start < end) remove_unix_locks( fd, start, end + 1 );
929 /* add a lock on an fd */
930 /* returns handle to wait on */
931 obj_handle_t lock_fd( struct fd *fd, file_pos_t start, file_pos_t count, int shared, int wait )
934 file_pos_t end = start + count;
936 /* don't allow wrapping locks */
937 if (end && end < start)
939 set_error( STATUS_INVALID_PARAMETER );
943 /* check if another lock on that file overlaps the area */
944 LIST_FOR_EACH( ptr, &fd->inode->locks )
946 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
947 if (!lock_overlaps( lock, start, end )) continue;
948 if (lock->shared && shared) continue;
952 set_error( STATUS_FILE_LOCK_CONFLICT );
955 set_error( STATUS_PENDING );
956 return alloc_handle( current->process, lock, SYNCHRONIZE, 0 );
959 /* not found, add it */
960 if (add_lock( fd, shared, start, end )) return 0;
961 if (get_error() == STATUS_FILE_LOCK_CONFLICT)
963 /* Unix lock conflict -> tell client to wait and retry */
964 if (wait) set_error( STATUS_PENDING );
969 /* remove a lock on an fd */
970 void unlock_fd( struct fd *fd, file_pos_t start, file_pos_t count )
973 file_pos_t end = start + count;
975 /* find an existing lock with the exact same parameters */
976 LIST_FOR_EACH( ptr, &fd->locks )
978 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
979 if ((lock->start == start) && (lock->end == end))
981 remove_lock( lock, 1 );
985 set_error( STATUS_FILE_LOCK_CONFLICT );
989 /****************************************************************/
990 /* asynchronous operations support */
994 struct thread *thread;
998 struct timeout_user *timeout;
1002 /* notifies client thread of new status of its async request */
1003 /* destroys the server side of it */
1004 static void async_terminate( struct async *async, int status )
1006 thread_queue_apc( async->thread, NULL, async->apc, APC_ASYNC_IO,
1007 1, async->user, async->sb, (void *)status );
1009 if (async->timeout) remove_timeout_user( async->timeout );
1010 async->timeout = NULL;
1011 list_remove( &async->entry );
1012 release_object( async->thread );
1016 /* cb for timeout on an async request */
1017 static void async_callback(void *private)
1019 struct async *async = (struct async *)private;
1021 /* fprintf(stderr, "async timeout out %p\n", async); */
1022 async->timeout = NULL;
1023 async_terminate( async, STATUS_TIMEOUT );
1026 /* create an async on a given queue of a fd */
1027 struct async *create_async(struct thread *thread, int* timeout, struct list *queue,
1028 void *io_apc, void *io_user, void* io_sb)
1030 struct async *async = mem_alloc( sizeof(struct async) );
1032 if (!async) return NULL;
1034 async->thread = (struct thread *)grab_object(thread);
1035 async->apc = io_apc;
1036 async->user = io_user;
1039 list_add_tail( queue, &async->entry );
1043 struct timeval when;
1045 gettimeofday( &when, 0 );
1046 add_timeout( &when, *timeout );
1047 async->timeout = add_timeout_user( &when, async_callback, async );
1049 else async->timeout = NULL;
1054 /* terminate the async operation at the head of the queue */
1055 void async_terminate_head( struct list *queue, int status )
1057 struct list *ptr = list_head( queue );
1058 if (ptr) async_terminate( LIST_ENTRY( ptr, struct async, entry ), status );
1061 /****************************************************************/
1062 /* file descriptor functions */
1064 static void fd_dump( struct object *obj, int verbose )
1066 struct fd *fd = (struct fd *)obj;
1067 fprintf( stderr, "Fd unix_fd=%d user=%p", fd->unix_fd, fd->user );
1068 if (fd->inode) fprintf( stderr, " inode=%p unlink='%s'", fd->inode, fd->closed->unlink );
1069 fprintf( stderr, "\n" );
1072 static void fd_destroy( struct object *obj )
1074 struct fd *fd = (struct fd *)obj;
1076 remove_fd_locks( fd );
1077 list_remove( &fd->inode_entry );
1078 if (fd->poll_index != -1) remove_poll_user( fd, fd->poll_index );
1081 inode_add_closed_fd( fd->inode, fd->closed );
1082 release_object( fd->inode );
1084 else /* no inode, close it right away */
1086 if (fd->unix_fd != -1) close( fd->unix_fd );
1090 /* set the events that select waits for on this fd */
1091 void set_fd_events( struct fd *fd, int events )
1093 int user = fd->poll_index;
1094 assert( poll_users[user] == fd );
1096 set_fd_epoll_events( fd, user, events );
1098 if (events == -1) /* stop waiting on this fd completely */
1100 pollfd[user].fd = -1;
1101 pollfd[user].events = POLLERR;
1102 pollfd[user].revents = 0;
1104 else if (pollfd[user].fd != -1 || !pollfd[user].events)
1106 pollfd[user].fd = fd->unix_fd;
1107 pollfd[user].events = events;
1111 /* allocate an fd object, without setting the unix fd yet */
1112 struct fd *alloc_fd( const struct fd_ops *fd_user_ops, struct object *user )
1114 struct fd *fd = alloc_object( &fd_ops );
1116 if (!fd) return NULL;
1118 fd->fd_ops = fd_user_ops;
1126 fd->poll_index = -1;
1127 list_init( &fd->inode_entry );
1128 list_init( &fd->locks );
1130 if ((fd->poll_index = add_poll_user( fd )) == -1)
1132 release_object( fd );
1138 /* check if the desired access is possible without violating */
1139 /* the sharing mode of other opens of the same file */
1140 static int check_sharing( struct fd *fd, unsigned int access, unsigned int sharing )
1142 unsigned int existing_sharing = FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
1143 unsigned int existing_access = 0;
1147 /* if access mode is 0, sharing mode is ignored */
1148 if (!access) sharing = existing_sharing;
1149 fd->access = access;
1150 fd->sharing = sharing;
1152 LIST_FOR_EACH( ptr, &fd->inode->open )
1154 struct fd *fd_ptr = LIST_ENTRY( ptr, struct fd, inode_entry );
1157 existing_sharing &= fd_ptr->sharing;
1158 existing_access |= fd_ptr->access;
1159 if (fd_ptr->closed->unlink[0]) unlink = 1;
1163 if ((access & GENERIC_READ) && !(existing_sharing & FILE_SHARE_READ)) return 0;
1164 if ((access & GENERIC_WRITE) && !(existing_sharing & FILE_SHARE_WRITE)) return 0;
1165 if ((existing_access & GENERIC_READ) && !(sharing & FILE_SHARE_READ)) return 0;
1166 if ((existing_access & GENERIC_WRITE) && !(sharing & FILE_SHARE_WRITE)) return 0;
1167 if (fd->closed->unlink[0] && !(existing_sharing & FILE_SHARE_DELETE)) return 0;
1168 if (unlink && !(sharing & FILE_SHARE_DELETE)) return 0;
1172 /* open() wrapper using a struct fd */
1173 /* the fd must have been created with alloc_fd */
1174 /* on error the fd object is released */
1175 struct fd *open_fd( struct fd *fd, const char *name, int flags, mode_t *mode,
1176 unsigned int access, unsigned int sharing, unsigned int options )
1179 struct closed_fd *closed_fd;
1180 const char *unlink_name = "";
1182 assert( fd->unix_fd == -1 );
1184 if (options & FILE_DELETE_ON_CLOSE) unlink_name = name;
1185 if (!(closed_fd = mem_alloc( sizeof(*closed_fd) + strlen(unlink_name) )))
1187 release_object( fd );
1190 /* create the directory if needed */
1191 if ((options & FILE_DIRECTORY_FILE) && (flags & O_CREAT))
1193 if (mkdir( name, 0777 ) == -1)
1195 if (errno != EEXIST || (flags & O_EXCL))
1198 release_object( fd );
1203 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
1205 if ((fd->unix_fd = open( name, flags & ~O_TRUNC, *mode )) == -1)
1208 release_object( fd );
1212 closed_fd->fd = fd->unix_fd;
1213 closed_fd->unlink[0] = 0;
1214 fstat( fd->unix_fd, &st );
1217 /* only bother with an inode for normal files and directories */
1218 if (S_ISREG(st.st_mode) || S_ISDIR(st.st_mode))
1220 struct inode *inode = get_inode( st.st_dev, st.st_ino );
1224 /* we can close the fd because there are no others open on the same file,
1225 * otherwise we wouldn't have failed to allocate a new inode
1230 fd->closed = closed_fd;
1231 list_add_head( &inode->open, &fd->inode_entry );
1233 /* check directory options */
1234 if ((options & FILE_DIRECTORY_FILE) && !S_ISDIR(st.st_mode))
1236 release_object( fd );
1237 set_error( STATUS_NOT_A_DIRECTORY );
1240 if ((options & FILE_NON_DIRECTORY_FILE) && S_ISDIR(st.st_mode))
1242 release_object( fd );
1243 set_error( STATUS_FILE_IS_A_DIRECTORY );
1246 if (!check_sharing( fd, access, sharing ))
1248 release_object( fd );
1249 set_error( STATUS_SHARING_VIOLATION );
1252 strcpy( closed_fd->unlink, unlink_name );
1253 if (flags & O_TRUNC) ftruncate( fd->unix_fd, 0 );
1255 else /* special file */
1257 if (options & FILE_DIRECTORY_FILE)
1259 set_error( STATUS_NOT_A_DIRECTORY );
1262 if (unlink_name[0]) /* we can't unlink special files */
1264 set_error( STATUS_INVALID_PARAMETER );
1272 release_object( fd );
1277 /* create an fd for an anonymous file */
1278 /* if the function fails the unix fd is closed */
1279 struct fd *create_anonymous_fd( const struct fd_ops *fd_user_ops, int unix_fd, struct object *user )
1281 struct fd *fd = alloc_fd( fd_user_ops, user );
1285 fd->unix_fd = unix_fd;
1292 /* retrieve the object that is using an fd */
1293 void *get_fd_user( struct fd *fd )
1298 /* retrieve the unix fd for an object */
1299 int get_unix_fd( struct fd *fd )
1304 /* check if two file descriptors point to the same file */
1305 int is_same_file_fd( struct fd *fd1, struct fd *fd2 )
1307 return fd1->inode == fd2->inode;
1310 /* callback for event happening in the main poll() loop */
1311 void fd_poll_event( struct fd *fd, int event )
1313 return fd->fd_ops->poll_event( fd, event );
1316 /* check if events are pending and if yes return which one(s) */
1317 int check_fd_events( struct fd *fd, int events )
1321 pfd.fd = fd->unix_fd;
1322 pfd.events = events;
1323 if (poll( &pfd, 1, 0 ) <= 0) return 0;
1327 /* default add_queue() routine for objects that poll() on an fd */
1328 int default_fd_add_queue( struct object *obj, struct wait_queue_entry *entry )
1330 struct fd *fd = get_obj_fd( obj );
1333 if (list_empty( &obj->wait_queue )) /* first on the queue */
1334 set_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1335 add_queue( obj, entry );
1336 release_object( fd );
1340 /* default remove_queue() routine for objects that poll() on an fd */
1341 void default_fd_remove_queue( struct object *obj, struct wait_queue_entry *entry )
1343 struct fd *fd = get_obj_fd( obj );
1346 remove_queue( obj, entry );
1347 if (list_empty( &obj->wait_queue )) /* last on the queue is gone */
1348 set_fd_events( fd, 0 );
1349 release_object( obj );
1350 release_object( fd );
1353 /* default signaled() routine for objects that poll() on an fd */
1354 int default_fd_signaled( struct object *obj, struct thread *thread )
1357 struct fd *fd = get_obj_fd( obj );
1359 if (fd->inode) return 1; /* regular files are always signaled */
1361 events = fd->fd_ops->get_poll_events( fd );
1362 ret = check_fd_events( fd, events ) != 0;
1365 set_fd_events( fd, 0 ); /* stop waiting on select() if we are signaled */
1366 else if (!list_empty( &obj->wait_queue ))
1367 set_fd_events( fd, events ); /* restart waiting on poll() if we are no longer signaled */
1369 release_object( fd );
1373 /* default handler for poll() events */
1374 void default_poll_event( struct fd *fd, int event )
1376 /* an error occurred, stop polling this fd to avoid busy-looping */
1377 if (event & (POLLERR | POLLHUP)) set_fd_events( fd, -1 );
1378 wake_up( fd->user, 0 );
1381 /* default flush() routine */
1382 int no_flush( struct fd *fd, struct event **event )
1384 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1388 /* default get_file_info() routine */
1389 int no_get_file_info( struct fd *fd )
1391 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1395 /* default queue_async() routine */
1396 void no_queue_async( struct fd *fd, void* apc, void* user, void* io_sb,
1397 int type, int count)
1399 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1402 /* default cancel_async() routine */
1403 void no_cancel_async( struct fd *fd )
1405 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1408 /* same as get_handle_obj but retrieve the struct fd associated to the object */
1409 static struct fd *get_handle_fd_obj( struct process *process, obj_handle_t handle,
1410 unsigned int access )
1412 struct fd *fd = NULL;
1415 if ((obj = get_handle_obj( process, handle, access, NULL )))
1417 fd = get_obj_fd( obj );
1418 release_object( obj );
1423 /* flush a file buffers */
1424 DECL_HANDLER(flush_file)
1426 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1427 struct event * event = NULL;
1431 fd->fd_ops->flush( fd, &event );
1434 reply->event = alloc_handle( current->process, event, SYNCHRONIZE, 0 );
1436 release_object( fd );
1440 /* get a Unix fd to access a file */
1441 DECL_HANDLER(get_handle_fd)
1447 if ((fd = get_handle_fd_obj( current->process, req->handle, req->access )))
1449 int unix_fd = get_handle_unix_fd( current->process, req->handle, req->access );
1450 if (unix_fd != -1) reply->fd = unix_fd;
1451 else if (!get_error())
1453 assert( fd->unix_fd != -1 );
1454 send_client_fd( current->process, fd->unix_fd, req->handle );
1456 reply->flags = fd->fd_ops->get_file_info( fd );
1457 release_object( fd );
1461 /* create / reschedule an async I/O */
1462 DECL_HANDLER(register_async)
1464 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1467 * The queue_async method must do the following:
1469 * 1. Get the async_queue for the request of given type.
1470 * 2. Create a new asynchronous request for the selected queue
1471 * 3. Carry out any operations necessary to adjust the object's poll events
1472 * Usually: set_elect_events (obj, obj->ops->get_poll_events()).
1473 * 4. When the async request is triggered, then send back (with a proper APC)
1474 * the trigger (STATUS_ALERTED) to the thread that posted the request.
1475 * async_destroy() is to be called: it will both notify the sender about
1476 * the trigger and destroy the request by itself
1477 * See also the implementations in file.c, serial.c, and sock.c.
1482 fd->fd_ops->queue_async( fd, req->io_apc, req->io_user, req->io_sb,
1483 req->type, req->count );
1484 release_object( fd );
1488 /* cancels all async I/O */
1489 DECL_HANDLER(cancel_async)
1491 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1494 /* Note: we don't kill the queued APC_ASYNC_IO on this thread because
1495 * NtCancelIoFile() will force the pending APC to be run. Since,
1496 * Windows only guarantees that the current thread will have no async
1497 * operation on the current fd when NtCancelIoFile returns, this shall
1500 fd->fd_ops->cancel_async( fd );
1501 release_object( fd );