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
23 #include "wine/port.h"
37 #ifdef HAVE_SYS_POLL_H
45 #include <sys/types.h>
49 #define WIN32_NO_STATUS
58 #if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
59 # include <sys/epoll.h>
61 #elif defined(linux) && defined(__i386__) && defined(HAVE_STDINT_H)
63 # define EPOLLIN POLLIN
64 # define EPOLLOUT POLLOUT
65 # define EPOLLERR POLLERR
66 # define EPOLLHUP POLLHUP
67 # define EPOLL_CTL_ADD 1
68 # define EPOLL_CTL_DEL 2
69 # define EPOLL_CTL_MOD 3
71 typedef union epoll_data
85 #define SYSCALL_RET(ret) do { \
86 if (ret < 0) { errno = -ret; ret = -1; } \
90 static inline int epoll_create( int size )
93 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
94 : "=a" (ret) : "0" (254 /*NR_epoll_create*/), "r" (size) );
98 static inline int epoll_ctl( int epfd, int op, int fd, const struct epoll_event *event )
101 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
103 : "0" (255 /*NR_epoll_ctl*/), "r" (epfd), "c" (op), "d" (fd), "S" (event), "m" (*event) );
107 static inline int epoll_wait( int epfd, struct epoll_event *events, int maxevents, int timeout )
110 __asm__( "pushl %%ebx; movl %2,%%ebx; int $0x80; popl %%ebx"
112 : "0" (256 /*NR_epoll_wait*/), "r" (epfd), "c" (events), "d" (maxevents), "S" (timeout)
118 #endif /* linux && __i386__ && HAVE_STDINT_H */
121 /* Because of the stupid Posix locking semantics, we need to keep
122 * track of all file descriptors referencing a given file, and not
123 * close a single one until all the locks are gone (sigh).
126 /* file descriptor object */
128 /* closed_fd is used to keep track of the unix fd belonging to a closed fd object */
131 struct list entry; /* entry in inode closed list */
132 int unix_fd; /* the unix file descriptor */
133 char unlink[1]; /* name to unlink on close (if any) */
138 struct object obj; /* object header */
139 const struct fd_ops *fd_ops; /* file descriptor operations */
140 struct inode *inode; /* inode that this fd belongs to */
141 struct list inode_entry; /* entry in inode fd list */
142 struct closed_fd *closed; /* structure to store the unix fd at destroy time */
143 struct object *user; /* object using this file descriptor */
144 struct list locks; /* list of locks on this fd */
145 unsigned int access; /* file access (FILE_READ_DATA etc.) */
146 unsigned int sharing; /* file sharing mode */
147 int unix_fd; /* unix file descriptor */
148 int fs_locks :1; /* can we use filesystem locks for this fd? */
149 int unmounted :1;/* has the device been unmounted? */
150 int poll_index; /* index of fd in poll array */
151 struct list read_q; /* async readers of this fd */
152 struct list write_q; /* async writers of this fd */
155 static void fd_dump( struct object *obj, int verbose );
156 static void fd_destroy( struct object *obj );
158 static const struct object_ops fd_ops =
160 sizeof(struct fd), /* size */
162 no_add_queue, /* add_queue */
163 NULL, /* remove_queue */
165 NULL, /* satisfied */
166 no_signal, /* signal */
167 no_get_fd, /* get_fd */
168 no_map_access, /* map_access */
169 no_lookup_name, /* lookup_name */
170 no_close_handle, /* close_handle */
171 fd_destroy /* destroy */
176 #define DEVICE_HASH_SIZE 7
177 #define INODE_HASH_SIZE 17
181 struct object obj; /* object header */
182 struct list entry; /* entry in device hash list */
183 dev_t dev; /* device number */
184 int removable; /* removable device? (or -1 if unknown) */
185 struct list inode_hash[INODE_HASH_SIZE]; /* inodes hash table */
188 static void device_dump( struct object *obj, int verbose );
189 static void device_destroy( struct object *obj );
191 static const struct object_ops device_ops =
193 sizeof(struct device), /* size */
194 device_dump, /* dump */
195 no_add_queue, /* add_queue */
196 NULL, /* remove_queue */
198 NULL, /* satisfied */
199 no_signal, /* signal */
200 no_get_fd, /* get_fd */
201 no_map_access, /* map_access */
202 no_lookup_name, /* lookup_name */
203 no_close_handle, /* close_handle */
204 device_destroy /* destroy */
211 struct object obj; /* object header */
212 struct list entry; /* inode hash list entry */
213 struct device *device; /* device containing this inode */
214 ino_t ino; /* inode number */
215 struct list open; /* list of open file descriptors */
216 struct list locks; /* list of file locks */
217 struct list closed; /* list of file descriptors to close at destroy time */
220 static void inode_dump( struct object *obj, int verbose );
221 static void inode_destroy( struct object *obj );
223 static const struct object_ops inode_ops =
225 sizeof(struct inode), /* size */
226 inode_dump, /* dump */
227 no_add_queue, /* add_queue */
228 NULL, /* remove_queue */
230 NULL, /* satisfied */
231 no_signal, /* signal */
232 no_get_fd, /* get_fd */
233 no_map_access, /* map_access */
234 no_lookup_name, /* lookup_name */
235 no_close_handle, /* close_handle */
236 inode_destroy /* destroy */
239 /* file lock object */
243 struct object obj; /* object header */
244 struct fd *fd; /* fd owning this lock */
245 struct list fd_entry; /* entry in list of locks on a given fd */
246 struct list inode_entry; /* entry in inode list of locks */
247 int shared; /* shared lock? */
248 file_pos_t start; /* locked region is interval [start;end) */
250 struct process *process; /* process owning this lock */
251 struct list proc_entry; /* entry in list of locks owned by the process */
254 static void file_lock_dump( struct object *obj, int verbose );
255 static int file_lock_signaled( struct object *obj, struct thread *thread );
257 static const struct object_ops file_lock_ops =
259 sizeof(struct file_lock), /* size */
260 file_lock_dump, /* dump */
261 add_queue, /* add_queue */
262 remove_queue, /* remove_queue */
263 file_lock_signaled, /* signaled */
264 no_satisfied, /* satisfied */
265 no_signal, /* signal */
266 no_get_fd, /* get_fd */
267 no_map_access, /* map_access */
268 no_lookup_name, /* lookup_name */
269 no_close_handle, /* close_handle */
270 no_destroy /* destroy */
274 #define OFF_T_MAX (~((file_pos_t)1 << (8*sizeof(off_t)-1)))
275 #define FILE_POS_T_MAX (~(file_pos_t)0)
277 static file_pos_t max_unix_offset = OFF_T_MAX;
279 #define DUMP_LONG_LONG(val) do { \
280 if (sizeof(val) > sizeof(unsigned long) && (val) > ~0UL) \
281 fprintf( stderr, "%lx%08lx", (unsigned long)((val) >> 32), (unsigned long)(val) ); \
283 fprintf( stderr, "%lx", (unsigned long)(val) ); \
288 /****************************************************************/
289 /* timeouts support */
293 struct list entry; /* entry in sorted timeout list */
294 struct timeval when; /* timeout expiry (absolute time) */
295 timeout_callback callback; /* callback function */
296 void *private; /* callback private data */
299 static struct list timeout_list = LIST_INIT(timeout_list); /* sorted timeouts list */
301 /* add a timeout user */
302 struct timeout_user *add_timeout_user( const struct timeval *when, timeout_callback func,
305 struct timeout_user *user;
308 if (!(user = mem_alloc( sizeof(*user) ))) return NULL;
310 user->callback = func;
311 user->private = private;
313 /* Now insert it in the linked list */
315 LIST_FOR_EACH( ptr, &timeout_list )
317 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
318 if (!time_before( &timeout->when, when )) break;
320 list_add_before( ptr, &user->entry );
324 /* remove a timeout user */
325 void remove_timeout_user( struct timeout_user *user )
327 list_remove( &user->entry );
331 /* add a timeout in milliseconds to an absolute time */
332 void add_timeout( struct timeval *when, int timeout )
336 long sec = timeout / 1000;
337 if ((when->tv_usec += (timeout - 1000*sec) * 1000) >= 1000000)
339 when->tv_usec -= 1000000;
347 /****************************************************************/
350 static struct fd **poll_users; /* users array */
351 static struct pollfd *pollfd; /* poll fd array */
352 static int nb_users; /* count of array entries actually in use */
353 static int active_users; /* current number of active users */
354 static int allocated_users; /* count of allocated entries in the array */
355 static struct fd **freelist; /* list of free entries in the array */
360 static struct epoll_event *epoll_events;
362 /* set the events that epoll waits for on this fd; helper for set_fd_events */
363 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
365 struct epoll_event ev;
368 if (epoll_fd == -1) return;
370 if (events == -1) /* stop waiting on this fd completely */
372 if (pollfd[user].fd == -1) return; /* already removed */
375 else if (pollfd[user].fd == -1)
377 if (pollfd[user].events) return; /* stopped waiting on it, don't restart */
382 if (pollfd[user].events == events) return; /* nothing to do */
387 memset(&ev.data, 0, sizeof(ev.data));
390 if (epoll_ctl( epoll_fd, ctl, fd->unix_fd, &ev ) == -1)
392 if (errno == ENOMEM) /* not enough memory, give up on epoll */
397 else perror( "epoll_ctl" ); /* should not happen */
401 #else /* USE_EPOLL */
403 static inline void set_fd_epoll_events( struct fd *fd, int user, int events )
407 #endif /* USE_EPOLL */
410 /* add a user in the poll array and return its index, or -1 on failure */
411 static int add_poll_user( struct fd *fd )
416 ret = freelist - poll_users;
417 freelist = (struct fd **)poll_users[ret];
421 if (nb_users == allocated_users)
423 struct fd **newusers;
424 struct pollfd *newpoll;
425 int new_count = allocated_users ? (allocated_users + allocated_users / 2) : 16;
426 if (!(newusers = realloc( poll_users, new_count * sizeof(*poll_users) ))) return -1;
427 if (!(newpoll = realloc( pollfd, new_count * sizeof(*pollfd) )))
430 poll_users = newusers;
435 poll_users = newusers;
438 if (!allocated_users) epoll_fd = epoll_create( new_count );
441 struct epoll_event *new_events;
442 if (!(new_events = realloc( epoll_events, new_count * sizeof(*epoll_events) )))
444 epoll_events = new_events;
447 allocated_users = new_count;
452 pollfd[ret].events = 0;
453 pollfd[ret].revents = 0;
454 poll_users[ret] = fd;
459 /* remove a user from the poll list */
460 static void remove_poll_user( struct fd *fd, int user )
463 assert( poll_users[user] == fd );
466 if (epoll_fd != -1 && pollfd[user].fd != -1)
468 struct epoll_event dummy;
469 epoll_ctl( epoll_fd, EPOLL_CTL_DEL, fd->unix_fd, &dummy );
472 pollfd[user].fd = -1;
473 pollfd[user].events = 0;
474 pollfd[user].revents = 0;
475 poll_users[user] = (struct fd *)freelist;
476 freelist = &poll_users[user];
480 /* process pending timeouts and return the time until the next timeout, in milliseconds */
481 static int get_next_timeout(void)
483 if (!list_empty( &timeout_list ))
485 struct list expired_list, *ptr;
488 gettimeofday( &now, NULL );
490 /* first remove all expired timers from the list */
492 list_init( &expired_list );
493 while ((ptr = list_head( &timeout_list )) != NULL)
495 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
497 if (!time_before( &now, &timeout->when ))
499 list_remove( &timeout->entry );
500 list_add_tail( &expired_list, &timeout->entry );
505 /* now call the callback for all the removed timers */
507 while ((ptr = list_head( &expired_list )) != NULL)
509 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
510 list_remove( &timeout->entry );
511 timeout->callback( timeout->private );
515 if ((ptr = list_head( &timeout_list )) != NULL)
517 struct timeout_user *timeout = LIST_ENTRY( ptr, struct timeout_user, entry );
518 int diff = (timeout->when.tv_sec - now.tv_sec) * 1000
519 + (timeout->when.tv_usec - now.tv_usec) / 1000;
520 if (diff < 0) diff = 0;
524 return -1; /* no pending timeouts */
527 /* server main poll() loop */
533 assert( POLLIN == EPOLLIN );
534 assert( POLLOUT == EPOLLOUT );
535 assert( POLLERR == EPOLLERR );
536 assert( POLLHUP == EPOLLHUP );
542 timeout = get_next_timeout();
544 if (!active_users) break; /* last user removed by a timeout */
545 if (epoll_fd == -1) break; /* an error occurred with epoll */
547 ret = epoll_wait( epoll_fd, epoll_events, allocated_users, timeout );
549 /* put the events into the pollfd array first, like poll does */
550 for (i = 0; i < ret; i++)
552 int user = epoll_events[i].data.u32;
553 pollfd[user].revents = epoll_events[i].events;
556 /* read events from the pollfd array, as set_fd_events may modify them */
557 for (i = 0; i < ret; i++)
559 int user = epoll_events[i].data.u32;
560 if (pollfd[user].revents) fd_poll_event( poll_users[user], pollfd[user].revents );
564 /* fall through to normal poll loop */
565 #endif /* USE_EPOLL */
569 timeout = get_next_timeout();
571 if (!active_users) break; /* last user removed by a timeout */
573 ret = poll( pollfd, nb_users, timeout );
576 for (i = 0; i < nb_users; i++)
578 if (pollfd[i].revents)
580 fd_poll_event( poll_users[i], pollfd[i].revents );
589 /****************************************************************/
590 /* device functions */
592 static struct list device_hash[DEVICE_HASH_SIZE];
594 /* retrieve the device object for a given fd, creating it if needed */
595 static struct device *get_device( dev_t dev, int create )
597 struct device *device;
598 unsigned int i, hash = dev % DEVICE_HASH_SIZE;
600 if (device_hash[hash].next)
602 LIST_FOR_EACH_ENTRY( device, &device_hash[hash], struct device, entry )
603 if (device->dev == dev) return (struct device *)grab_object( device );
605 else list_init( &device_hash[hash] );
607 /* not found, create it */
609 if (!create) return NULL;
610 if ((device = alloc_object( &device_ops )))
613 device->removable = -1;
614 for (i = 0; i < INODE_HASH_SIZE; i++) list_init( &device->inode_hash[i] );
615 list_add_head( &device_hash[hash], &device->entry );
620 static void device_dump( struct object *obj, int verbose )
622 struct device *device = (struct device *)obj;
623 fprintf( stderr, "Device dev=" );
624 DUMP_LONG_LONG( device->dev );
625 fprintf( stderr, "\n" );
628 static void device_destroy( struct object *obj )
630 struct device *device = (struct device *)obj;
633 for (i = 0; i < INODE_HASH_SIZE; i++)
634 assert( list_empty(&device->inode_hash[i]) );
636 list_remove( &device->entry ); /* remove it from the hash table */
640 /****************************************************************/
641 /* inode functions */
643 /* close all pending file descriptors in the closed list */
644 static void inode_close_pending( struct inode *inode, int keep_unlinks )
646 struct list *ptr = list_head( &inode->closed );
650 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
651 struct list *next = list_next( &inode->closed, ptr );
653 if (fd->unix_fd != -1)
655 close( fd->unix_fd );
658 if (!keep_unlinks || !fd->unlink[0]) /* get rid of it unless there's an unlink pending on that file */
667 static void inode_dump( struct object *obj, int verbose )
669 struct inode *inode = (struct inode *)obj;
670 fprintf( stderr, "Inode device=%p ino=", inode->device );
671 DUMP_LONG_LONG( inode->ino );
672 fprintf( stderr, "\n" );
675 static void inode_destroy( struct object *obj )
677 struct inode *inode = (struct inode *)obj;
680 assert( list_empty(&inode->open) );
681 assert( list_empty(&inode->locks) );
683 list_remove( &inode->entry );
685 while ((ptr = list_head( &inode->closed )))
687 struct closed_fd *fd = LIST_ENTRY( ptr, struct closed_fd, entry );
689 if (fd->unix_fd != -1) close( fd->unix_fd );
692 /* make sure it is still the same file */
694 if (!stat( fd->unlink, &st ) && st.st_dev == inode->device->dev && st.st_ino == inode->ino)
696 if (S_ISDIR(st.st_mode)) rmdir( fd->unlink );
697 else unlink( fd->unlink );
702 release_object( inode->device );
705 /* retrieve the inode object for a given fd, creating it if needed */
706 static struct inode *get_inode( dev_t dev, ino_t ino )
708 struct device *device;
710 unsigned int hash = ino % INODE_HASH_SIZE;
712 if (!(device = get_device( dev, 1 ))) return NULL;
714 LIST_FOR_EACH_ENTRY( inode, &device->inode_hash[hash], struct inode, entry )
716 if (inode->ino == ino)
718 release_object( device );
719 return (struct inode *)grab_object( inode );
723 /* not found, create it */
724 if ((inode = alloc_object( &inode_ops )))
726 inode->device = device;
728 list_init( &inode->open );
729 list_init( &inode->locks );
730 list_init( &inode->closed );
731 list_add_head( &device->inode_hash[hash], &inode->entry );
733 else release_object( device );
738 /* add fd to the inode list of file descriptors to close */
739 static void inode_add_closed_fd( struct inode *inode, struct closed_fd *fd )
741 if (!list_empty( &inode->locks ))
743 list_add_head( &inode->closed, &fd->entry );
745 else if (fd->unlink[0]) /* close the fd but keep the structure around for unlink */
747 if (fd->unix_fd != -1) close( fd->unix_fd );
749 list_add_head( &inode->closed, &fd->entry );
751 else /* no locks on this inode and no unlink, get rid of the fd */
753 if (fd->unix_fd != -1) close( fd->unix_fd );
759 /****************************************************************/
760 /* file lock functions */
762 static void file_lock_dump( struct object *obj, int verbose )
764 struct file_lock *lock = (struct file_lock *)obj;
765 fprintf( stderr, "Lock %s fd=%p proc=%p start=",
766 lock->shared ? "shared" : "excl", lock->fd, lock->process );
767 DUMP_LONG_LONG( lock->start );
768 fprintf( stderr, " end=" );
769 DUMP_LONG_LONG( lock->end );
770 fprintf( stderr, "\n" );
773 static int file_lock_signaled( struct object *obj, struct thread *thread )
775 struct file_lock *lock = (struct file_lock *)obj;
776 /* lock is signaled if it has lost its owner */
777 return !lock->process;
780 /* set (or remove) a Unix lock if possible for the given range */
781 static int set_unix_lock( struct fd *fd, file_pos_t start, file_pos_t end, int type )
785 if (!fd->fs_locks) return 1; /* no fs locks possible for this fd */
788 if (start == end) return 1; /* can't set zero-byte lock */
789 if (start > max_unix_offset) return 1; /* ignore it */
791 fl.l_whence = SEEK_SET;
793 if (!end || end > max_unix_offset) fl.l_len = 0;
794 else fl.l_len = end - start;
795 if (fcntl( fd->unix_fd, F_SETLK, &fl ) != -1) return 1;
800 /* check whether locks work at all on this file system */
801 if (fcntl( fd->unix_fd, F_GETLK, &fl ) != -1)
803 set_error( STATUS_FILE_LOCK_CONFLICT );
809 /* no locking on this fs, just ignore it */
813 set_error( STATUS_FILE_LOCK_CONFLICT );
816 /* this can happen if we try to set a write lock on a read-only file */
817 /* we just ignore that error */
818 if (fl.l_type == F_WRLCK) return 1;
819 set_error( STATUS_ACCESS_DENIED );
825 /* this can happen if off_t is 64-bit but the kernel only supports 32-bit */
826 /* in that case we shrink the limit and retry */
827 if (max_unix_offset > INT_MAX)
829 max_unix_offset = INT_MAX;
840 /* check if interval [start;end) overlaps the lock */
841 inline static int lock_overlaps( struct file_lock *lock, file_pos_t start, file_pos_t end )
843 if (lock->end && start >= lock->end) return 0;
844 if (end && lock->start >= end) return 0;
848 /* remove Unix locks for all bytes in the specified area that are no longer locked */
849 static void remove_unix_locks( struct fd *fd, file_pos_t start, file_pos_t end )
857 } *first, *cur, *next, *buffer;
862 if (!fd->inode) return;
863 if (!fd->fs_locks) return;
864 if (start == end || start > max_unix_offset) return;
865 if (!end || end > max_unix_offset) end = max_unix_offset + 1;
867 /* count the number of locks overlapping the specified area */
869 LIST_FOR_EACH( ptr, &fd->inode->locks )
871 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
872 if (lock->start == lock->end) continue;
873 if (lock_overlaps( lock, start, end )) count++;
876 if (!count) /* no locks at all, we can unlock everything */
878 set_unix_lock( fd, start, end, F_UNLCK );
882 /* allocate space for the list of holes */
883 /* max. number of holes is number of locks + 1 */
885 if (!(buffer = malloc( sizeof(*buffer) * (count+1) ))) return;
889 first->start = start;
893 /* build a sorted list of unlocked holes in the specified area */
895 LIST_FOR_EACH( ptr, &fd->inode->locks )
897 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
898 if (lock->start == lock->end) continue;
899 if (!lock_overlaps( lock, start, end )) continue;
901 /* go through all the holes touched by this lock */
902 for (cur = first; cur; cur = cur->next)
904 if (cur->end <= lock->start) continue; /* hole is before start of lock */
905 if (lock->end && cur->start >= lock->end) break; /* hole is after end of lock */
907 /* now we know that lock is overlapping hole */
909 if (cur->start >= lock->start) /* lock starts before hole, shrink from start */
911 cur->start = lock->end;
912 if (cur->start && cur->start < cur->end) break; /* done with this lock */
913 /* now hole is empty, remove it */
914 if (cur->next) cur->next->prev = cur->prev;
915 if (cur->prev) cur->prev->next = cur->next;
916 else if (!(first = cur->next)) goto done; /* no more holes at all */
918 else if (!lock->end || cur->end <= lock->end) /* lock larger than hole, shrink from end */
920 cur->end = lock->start;
921 assert( cur->start < cur->end );
923 else /* lock is in the middle of hole, split hole in two */
926 next->next = cur->next;
928 next->start = lock->end;
929 next->end = cur->end;
930 cur->end = lock->start;
931 assert( next->start < next->end );
932 assert( cur->end < next->start );
934 break; /* done with this lock */
939 /* clear Unix locks for all the holes */
941 for (cur = first; cur; cur = cur->next)
942 set_unix_lock( fd, cur->start, cur->end, F_UNLCK );
948 /* create a new lock on a fd */
949 static struct file_lock *add_lock( struct fd *fd, int shared, file_pos_t start, file_pos_t end )
951 struct file_lock *lock;
953 if (!fd->inode) /* not a regular file */
955 set_error( STATUS_INVALID_HANDLE );
959 if (!(lock = alloc_object( &file_lock_ops ))) return NULL;
960 lock->shared = shared;
964 lock->process = current->process;
966 /* now try to set a Unix lock */
967 if (!set_unix_lock( lock->fd, lock->start, lock->end, lock->shared ? F_RDLCK : F_WRLCK ))
969 release_object( lock );
972 list_add_head( &fd->locks, &lock->fd_entry );
973 list_add_head( &fd->inode->locks, &lock->inode_entry );
974 list_add_head( &lock->process->locks, &lock->proc_entry );
978 /* remove an existing lock */
979 static void remove_lock( struct file_lock *lock, int remove_unix )
981 struct inode *inode = lock->fd->inode;
983 list_remove( &lock->fd_entry );
984 list_remove( &lock->inode_entry );
985 list_remove( &lock->proc_entry );
986 if (remove_unix) remove_unix_locks( lock->fd, lock->start, lock->end );
987 if (list_empty( &inode->locks )) inode_close_pending( inode, 1 );
988 lock->process = NULL;
989 wake_up( &lock->obj, 0 );
990 release_object( lock );
993 /* remove all locks owned by a given process */
994 void remove_process_locks( struct process *process )
998 while ((ptr = list_head( &process->locks )))
1000 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, proc_entry );
1001 remove_lock( lock, 1 ); /* this removes it from the list */
1005 /* remove all locks on a given fd */
1006 static void remove_fd_locks( struct fd *fd )
1008 file_pos_t start = FILE_POS_T_MAX, end = 0;
1011 while ((ptr = list_head( &fd->locks )))
1013 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
1014 if (lock->start < start) start = lock->start;
1015 if (!lock->end || lock->end > end) end = lock->end - 1;
1016 remove_lock( lock, 0 );
1018 if (start < end) remove_unix_locks( fd, start, end + 1 );
1021 /* add a lock on an fd */
1022 /* returns handle to wait on */
1023 obj_handle_t lock_fd( struct fd *fd, file_pos_t start, file_pos_t count, int shared, int wait )
1026 file_pos_t end = start + count;
1028 /* don't allow wrapping locks */
1029 if (end && end < start)
1031 set_error( STATUS_INVALID_PARAMETER );
1035 /* check if another lock on that file overlaps the area */
1036 LIST_FOR_EACH( ptr, &fd->inode->locks )
1038 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, inode_entry );
1039 if (!lock_overlaps( lock, start, end )) continue;
1040 if (lock->shared && shared) continue;
1044 set_error( STATUS_FILE_LOCK_CONFLICT );
1047 set_error( STATUS_PENDING );
1048 return alloc_handle( current->process, lock, SYNCHRONIZE, 0 );
1051 /* not found, add it */
1052 if (add_lock( fd, shared, start, end )) return 0;
1053 if (get_error() == STATUS_FILE_LOCK_CONFLICT)
1055 /* Unix lock conflict -> tell client to wait and retry */
1056 if (wait) set_error( STATUS_PENDING );
1061 /* remove a lock on an fd */
1062 void unlock_fd( struct fd *fd, file_pos_t start, file_pos_t count )
1065 file_pos_t end = start + count;
1067 /* find an existing lock with the exact same parameters */
1068 LIST_FOR_EACH( ptr, &fd->locks )
1070 struct file_lock *lock = LIST_ENTRY( ptr, struct file_lock, fd_entry );
1071 if ((lock->start == start) && (lock->end == end))
1073 remove_lock( lock, 1 );
1077 set_error( STATUS_FILE_LOCK_CONFLICT );
1081 /****************************************************************/
1082 /* asynchronous operations support */
1086 struct thread *thread;
1090 struct timeout_user *timeout;
1094 /* notifies client thread of new status of its async request */
1095 /* destroys the server side of it */
1096 static void async_terminate( struct async *async, int status )
1098 thread_queue_apc( async->thread, NULL, async->apc, APC_ASYNC_IO,
1099 1, async->user, async->sb, (void *)status );
1101 if (async->timeout) remove_timeout_user( async->timeout );
1102 async->timeout = NULL;
1103 list_remove( &async->entry );
1104 release_object( async->thread );
1108 /* cb for timeout on an async request */
1109 static void async_callback(void *private)
1111 struct async *async = (struct async *)private;
1113 /* fprintf(stderr, "async timeout out %p\n", async); */
1114 async->timeout = NULL;
1115 async_terminate( async, STATUS_TIMEOUT );
1118 /* create an async on a given queue of a fd */
1119 struct async *create_async(struct thread *thread, int* timeout, struct list *queue,
1120 void *io_apc, void *io_user, void* io_sb)
1122 struct async *async = mem_alloc( sizeof(struct async) );
1124 if (!async) return NULL;
1126 async->thread = (struct thread *)grab_object(thread);
1127 async->apc = io_apc;
1128 async->user = io_user;
1131 list_add_tail( queue, &async->entry );
1135 struct timeval when;
1137 gettimeofday( &when, NULL );
1138 add_timeout( &when, *timeout );
1139 async->timeout = add_timeout_user( &when, async_callback, async );
1141 else async->timeout = NULL;
1146 /* terminate the async operation at the head of the queue */
1147 void async_terminate_head( struct list *queue, int status )
1149 struct list *ptr = list_head( queue );
1150 if (ptr) async_terminate( LIST_ENTRY( ptr, struct async, entry ), status );
1153 /****************************************************************/
1154 /* file descriptor functions */
1156 static void fd_dump( struct object *obj, int verbose )
1158 struct fd *fd = (struct fd *)obj;
1159 fprintf( stderr, "Fd unix_fd=%d user=%p", fd->unix_fd, fd->user );
1160 if (fd->inode) fprintf( stderr, " inode=%p unlink='%s'", fd->inode, fd->closed->unlink );
1161 fprintf( stderr, "\n" );
1164 static void fd_destroy( struct object *obj )
1166 struct fd *fd = (struct fd *)obj;
1168 async_terminate_queue( &fd->read_q, STATUS_CANCELLED );
1169 async_terminate_queue( &fd->write_q, STATUS_CANCELLED );
1171 remove_fd_locks( fd );
1172 list_remove( &fd->inode_entry );
1173 if (fd->poll_index != -1) remove_poll_user( fd, fd->poll_index );
1176 inode_add_closed_fd( fd->inode, fd->closed );
1177 release_object( fd->inode );
1179 else /* no inode, close it right away */
1181 if (fd->unix_fd != -1) close( fd->unix_fd );
1185 /* set the events that select waits for on this fd */
1186 void set_fd_events( struct fd *fd, int events )
1188 int user = fd->poll_index;
1189 assert( poll_users[user] == fd );
1191 set_fd_epoll_events( fd, user, events );
1193 if (events == -1) /* stop waiting on this fd completely */
1195 pollfd[user].fd = -1;
1196 pollfd[user].events = POLLERR;
1197 pollfd[user].revents = 0;
1199 else if (pollfd[user].fd != -1 || !pollfd[user].events)
1201 pollfd[user].fd = fd->unix_fd;
1202 pollfd[user].events = events;
1206 /* prepare an fd for unmounting its corresponding device */
1207 static inline void unmount_fd( struct fd *fd )
1209 assert( fd->inode );
1211 async_terminate_queue( &fd->read_q, STATUS_VOLUME_DISMOUNTED );
1212 async_terminate_queue( &fd->write_q, STATUS_VOLUME_DISMOUNTED );
1214 if (fd->poll_index != -1) set_fd_events( fd, -1 );
1216 if (fd->unix_fd != -1) close( fd->unix_fd );
1220 fd->closed->unix_fd = -1;
1221 fd->closed->unlink[0] = 0;
1223 /* stop using Unix locks on this fd (existing locks have been removed by close) */
1227 /* allocate an fd object, without setting the unix fd yet */
1228 static struct fd *alloc_fd_object(void)
1230 struct fd *fd = alloc_object( &fd_ops );
1232 if (!fd) return NULL;
1243 fd->poll_index = -1;
1244 list_init( &fd->inode_entry );
1245 list_init( &fd->locks );
1246 list_init( &fd->read_q );
1247 list_init( &fd->write_q );
1249 if ((fd->poll_index = add_poll_user( fd )) == -1)
1251 release_object( fd );
1257 /* allocate a pseudo fd object, for objects that need to behave like files but don't have a unix fd */
1258 struct fd *alloc_pseudo_fd( const struct fd_ops *fd_user_ops, struct object *user )
1260 struct fd *fd = alloc_object( &fd_ops );
1262 if (!fd) return NULL;
1264 fd->fd_ops = fd_user_ops;
1273 fd->poll_index = -1;
1274 list_init( &fd->inode_entry );
1275 list_init( &fd->locks );
1276 list_init( &fd->read_q );
1277 list_init( &fd->write_q );
1281 /* check if the desired access is possible without violating */
1282 /* the sharing mode of other opens of the same file */
1283 static int check_sharing( struct fd *fd, unsigned int access, unsigned int sharing )
1285 unsigned int existing_sharing = FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE;
1286 unsigned int existing_access = 0;
1289 /* if access mode is 0, sharing mode is ignored */
1290 if (!access) sharing = existing_sharing;
1291 fd->access = access;
1292 fd->sharing = sharing;
1294 LIST_FOR_EACH( ptr, &fd->inode->open )
1296 struct fd *fd_ptr = LIST_ENTRY( ptr, struct fd, inode_entry );
1299 existing_sharing &= fd_ptr->sharing;
1300 existing_access |= fd_ptr->access;
1304 if ((access & FILE_UNIX_READ_ACCESS) && !(existing_sharing & FILE_SHARE_READ)) return 0;
1305 if ((access & FILE_UNIX_WRITE_ACCESS) && !(existing_sharing & FILE_SHARE_WRITE)) return 0;
1306 if ((access & DELETE) && !(existing_sharing & FILE_SHARE_DELETE)) return 0;
1307 if ((existing_access & FILE_UNIX_READ_ACCESS) && !(sharing & FILE_SHARE_READ)) return 0;
1308 if ((existing_access & FILE_UNIX_WRITE_ACCESS) && !(sharing & FILE_SHARE_WRITE)) return 0;
1309 if ((existing_access & DELETE) && !(sharing & FILE_SHARE_DELETE)) return 0;
1313 /* sets the user of an fd that previously had no user */
1314 void set_fd_user( struct fd *fd, const struct fd_ops *user_ops, struct object *user )
1316 assert( fd->fd_ops == NULL );
1317 fd->fd_ops = user_ops;
1321 /* open() wrapper that returns a struct fd with no fd user set */
1322 struct fd *open_fd( const char *name, int flags, mode_t *mode, unsigned int access,
1323 unsigned int sharing, unsigned int options )
1326 struct closed_fd *closed_fd;
1328 const char *unlink_name = "";
1331 if ((options & FILE_DELETE_ON_CLOSE) && !(access & DELETE))
1333 set_error( STATUS_INVALID_PARAMETER );
1337 if (!(fd = alloc_fd_object())) return NULL;
1339 if (options & FILE_DELETE_ON_CLOSE) unlink_name = name;
1340 if (!(closed_fd = mem_alloc( sizeof(*closed_fd) + strlen(unlink_name) )))
1342 release_object( fd );
1346 /* create the directory if needed */
1347 if ((options & FILE_DIRECTORY_FILE) && (flags & O_CREAT))
1349 if (mkdir( name, 0777 ) == -1)
1351 if (errno != EEXIST || (flags & O_EXCL))
1357 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
1360 if ((access & FILE_UNIX_WRITE_ACCESS) && !(options & FILE_DIRECTORY_FILE))
1362 if (access & FILE_UNIX_READ_ACCESS) rw_mode = O_RDWR;
1363 else rw_mode = O_WRONLY;
1365 else rw_mode = O_RDONLY;
1367 if ((fd->unix_fd = open( name, rw_mode | (flags & ~O_TRUNC), *mode )) == -1)
1369 /* if we tried to open a directory for write access, retry read-only */
1370 if (errno != EISDIR ||
1371 !(access & FILE_UNIX_WRITE_ACCESS) ||
1372 (fd->unix_fd = open( name, O_RDONLY | (flags & ~O_TRUNC), *mode )) == -1)
1379 closed_fd->unix_fd = fd->unix_fd;
1380 closed_fd->unlink[0] = 0;
1381 fstat( fd->unix_fd, &st );
1384 /* only bother with an inode for normal files and directories */
1385 if (S_ISREG(st.st_mode) || S_ISDIR(st.st_mode))
1387 struct inode *inode = get_inode( st.st_dev, st.st_ino );
1391 /* we can close the fd because there are no others open on the same file,
1392 * otherwise we wouldn't have failed to allocate a new inode
1397 fd->closed = closed_fd;
1398 list_add_head( &inode->open, &fd->inode_entry );
1400 /* check directory options */
1401 if ((options & FILE_DIRECTORY_FILE) && !S_ISDIR(st.st_mode))
1403 release_object( fd );
1404 set_error( STATUS_NOT_A_DIRECTORY );
1407 if ((options & FILE_NON_DIRECTORY_FILE) && S_ISDIR(st.st_mode))
1409 release_object( fd );
1410 set_error( STATUS_FILE_IS_A_DIRECTORY );
1413 if (!check_sharing( fd, access, sharing ))
1415 release_object( fd );
1416 set_error( STATUS_SHARING_VIOLATION );
1419 strcpy( closed_fd->unlink, unlink_name );
1420 if (flags & O_TRUNC) ftruncate( fd->unix_fd, 0 );
1422 else /* special file */
1424 if (options & FILE_DIRECTORY_FILE)
1426 set_error( STATUS_NOT_A_DIRECTORY );
1429 if (unlink_name[0]) /* we can't unlink special files */
1431 set_error( STATUS_INVALID_PARAMETER );
1439 release_object( fd );
1444 /* create an fd for an anonymous file */
1445 /* if the function fails the unix fd is closed */
1446 struct fd *create_anonymous_fd( const struct fd_ops *fd_user_ops, int unix_fd, struct object *user )
1448 struct fd *fd = alloc_fd_object();
1452 set_fd_user( fd, fd_user_ops, user );
1453 fd->unix_fd = unix_fd;
1460 /* retrieve the object that is using an fd */
1461 void *get_fd_user( struct fd *fd )
1466 /* retrieve the unix fd for an object */
1467 int get_unix_fd( struct fd *fd )
1469 if (fd->unix_fd == -1)
1471 if (fd->unmounted) set_error( STATUS_VOLUME_DISMOUNTED );
1472 else set_error( STATUS_BAD_DEVICE_TYPE );
1477 /* check if two file descriptors point to the same file */
1478 int is_same_file_fd( struct fd *fd1, struct fd *fd2 )
1480 return fd1->inode == fd2->inode;
1483 /* callback for event happening in the main poll() loop */
1484 void fd_poll_event( struct fd *fd, int event )
1486 return fd->fd_ops->poll_event( fd, event );
1489 /* check if events are pending and if yes return which one(s) */
1490 int check_fd_events( struct fd *fd, int events )
1494 if (fd->unix_fd == -1) return POLLERR;
1496 pfd.fd = fd->unix_fd;
1497 pfd.events = events;
1498 if (poll( &pfd, 1, 0 ) <= 0) return 0;
1502 /* default add_queue() routine for objects that poll() on an fd */
1503 int default_fd_add_queue( struct object *obj, struct wait_queue_entry *entry )
1505 struct fd *fd = get_obj_fd( obj );
1508 if (!fd->inode && list_empty( &obj->wait_queue )) /* first on the queue */
1509 set_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1510 add_queue( obj, entry );
1511 release_object( fd );
1515 /* default remove_queue() routine for objects that poll() on an fd */
1516 void default_fd_remove_queue( struct object *obj, struct wait_queue_entry *entry )
1518 struct fd *fd = get_obj_fd( obj );
1521 remove_queue( obj, entry );
1522 if (!fd->inode && list_empty( &obj->wait_queue )) /* last on the queue is gone */
1523 set_fd_events( fd, 0 );
1524 release_object( obj );
1525 release_object( fd );
1528 /* default signaled() routine for objects that poll() on an fd */
1529 int default_fd_signaled( struct object *obj, struct thread *thread )
1532 struct fd *fd = get_obj_fd( obj );
1534 if (fd->inode) ret = 1; /* regular files are always signaled */
1537 events = fd->fd_ops->get_poll_events( fd );
1538 ret = check_fd_events( fd, events ) != 0;
1542 /* stop waiting on select() if we are signaled */
1543 set_fd_events( fd, 0 );
1545 else if (!list_empty( &obj->wait_queue ))
1547 /* restart waiting on poll() if we are no longer signaled */
1548 set_fd_events( fd, events );
1551 release_object( fd );
1555 int default_fd_get_poll_events( struct fd *fd )
1559 if (!list_empty( &fd->read_q ))
1561 if (!list_empty( &fd->write_q ))
1567 /* default handler for poll() events */
1568 void default_poll_event( struct fd *fd, int event )
1570 if (!list_empty( &fd->read_q ) && (POLLIN & event) )
1572 async_terminate_head( &fd->read_q, STATUS_ALERTED );
1575 if (!list_empty( &fd->write_q ) && (POLLOUT & event) )
1577 async_terminate_head( &fd->write_q, STATUS_ALERTED );
1581 /* if an error occurred, stop polling this fd to avoid busy-looping */
1582 if (event & (POLLERR | POLLHUP)) set_fd_events( fd, -1 );
1583 wake_up( fd->user, 0 );
1586 void fd_queue_async_timeout( struct fd *fd, void *apc, void *user, void *io_sb, int type, int count, int *timeout )
1591 if (!(fd->fd_ops->get_file_info( fd ) & (FD_FLAG_OVERLAPPED|FD_FLAG_TIMEOUT)))
1593 set_error( STATUS_INVALID_HANDLE );
1599 case ASYNC_TYPE_READ:
1600 queue = &fd->read_q;
1602 case ASYNC_TYPE_WRITE:
1603 queue = &fd->write_q;
1606 set_error( STATUS_INVALID_PARAMETER );
1610 if (!create_async( current, timeout, queue, apc, user, io_sb ))
1613 /* Check if the new pending request can be served immediately */
1614 events = check_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1615 if (events) fd->fd_ops->poll_event( fd, events );
1617 set_fd_events( fd, fd->fd_ops->get_poll_events( fd ) );
1620 void default_fd_queue_async( struct fd *fd, void *apc, void *user, void *io_sb, int type, int count )
1622 fd_queue_async_timeout( fd, apc, user, io_sb, type, count, NULL );
1625 void default_fd_cancel_async( struct fd *fd )
1627 async_terminate_queue( &fd->read_q, STATUS_CANCELLED );
1628 async_terminate_queue( &fd->write_q, STATUS_CANCELLED );
1631 /* default flush() routine */
1632 int no_flush( struct fd *fd, struct event **event )
1634 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1638 /* default get_file_info() routine */
1639 int no_get_file_info( struct fd *fd )
1641 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1645 /* default queue_async() routine */
1646 void no_queue_async( struct fd *fd, void* apc, void* user, void* io_sb,
1647 int type, int count)
1649 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1652 /* default cancel_async() routine */
1653 void no_cancel_async( struct fd *fd )
1655 set_error( STATUS_OBJECT_TYPE_MISMATCH );
1658 /* close all Unix file descriptors on a device to allow unmounting it */
1659 static void unmount_device( struct fd *device_fd )
1663 struct device *device;
1664 struct inode *inode;
1666 int unix_fd = get_unix_fd( device_fd );
1668 if (unix_fd == -1) return;
1670 if (fstat( unix_fd, &st ) == -1 || !S_ISBLK( st.st_mode ))
1672 set_error( STATUS_INVALID_PARAMETER );
1676 if (!(device = get_device( st.st_rdev, 0 ))) return;
1678 for (i = 0; i < INODE_HASH_SIZE; i++)
1680 LIST_FOR_EACH_ENTRY( inode, &device->inode_hash[i], struct inode, entry )
1682 LIST_FOR_EACH_ENTRY( fd, &inode->open, struct fd, inode_entry )
1686 inode_close_pending( inode, 0 );
1689 /* remove it from the hash table */
1690 list_remove( &device->entry );
1691 list_init( &device->entry );
1692 release_object( device );
1695 /* same as get_handle_obj but retrieve the struct fd associated to the object */
1696 static struct fd *get_handle_fd_obj( struct process *process, obj_handle_t handle,
1697 unsigned int access )
1699 struct fd *fd = NULL;
1702 if ((obj = get_handle_obj( process, handle, access, NULL )))
1704 fd = get_obj_fd( obj );
1705 release_object( obj );
1710 /* flush a file buffers */
1711 DECL_HANDLER(flush_file)
1713 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1714 struct event * event = NULL;
1718 fd->fd_ops->flush( fd, &event );
1721 reply->event = alloc_handle( current->process, event, SYNCHRONIZE, 0 );
1723 release_object( fd );
1727 /* open a file object */
1728 DECL_HANDLER(open_file_object)
1730 struct unicode_str name;
1731 struct directory *root = NULL;
1734 get_req_unicode_str( &name );
1735 if (req->rootdir && !(root = get_directory_obj( current->process, req->rootdir, 0 )))
1738 if ((obj = open_object_dir( root, &name, req->attributes, NULL )))
1740 /* make sure this is a valid file object */
1741 struct fd *fd = get_obj_fd( obj );
1744 reply->handle = alloc_handle( current->process, obj, req->access, req->attributes );
1745 release_object( fd );
1747 release_object( obj );
1750 if (root) release_object( root );
1753 /* get a Unix fd to access a file */
1754 DECL_HANDLER(get_handle_fd)
1760 if ((fd = get_handle_fd_obj( current->process, req->handle, req->access )))
1762 int unix_fd = get_unix_fd( fd );
1765 int cached_fd = get_handle_unix_fd( current->process, req->handle, req->access );
1766 if (cached_fd != -1) reply->fd = cached_fd;
1767 else if (!get_error()) send_client_fd( current->process, unix_fd, req->handle );
1769 if (fd->inode) reply->removable = fd->inode->device->removable;
1770 reply->flags = fd->fd_ops->get_file_info( fd );
1771 release_object( fd );
1775 /* set the cached file descriptor of a handle */
1776 DECL_HANDLER(set_handle_fd)
1781 if ((fd = get_handle_fd_obj( current->process, req->handle, 0 )))
1783 struct device *device = fd->inode ? fd->inode->device : NULL;
1785 if (device && device->removable == -1) device->removable = req->removable;
1787 /* only cache the fd on non-removable devices */
1788 if (!device || !device->removable)
1789 reply->cur_fd = set_handle_unix_fd( current->process, req->handle, req->fd );
1790 release_object( fd );
1794 /* get ready to unmount a Unix device */
1795 DECL_HANDLER(unmount_device)
1799 if ((fd = get_handle_fd_obj( current->process, req->handle, 0 )))
1801 unmount_device( fd );
1802 release_object( fd );
1806 /* create / reschedule an async I/O */
1807 DECL_HANDLER(register_async)
1809 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1812 * The queue_async method must do the following:
1814 * 1. Get the async_queue for the request of given type.
1815 * 2. Create a new asynchronous request for the selected queue
1816 * 3. Carry out any operations necessary to adjust the object's poll events
1817 * Usually: set_elect_events (obj, obj->ops->get_poll_events()).
1818 * 4. When the async request is triggered, then send back (with a proper APC)
1819 * the trigger (STATUS_ALERTED) to the thread that posted the request.
1820 * async_destroy() is to be called: it will both notify the sender about
1821 * the trigger and destroy the request by itself
1822 * See also the implementations in file.c, serial.c, and sock.c.
1827 fd->fd_ops->queue_async( fd, req->io_apc, req->io_user, req->io_sb,
1828 req->type, req->count );
1829 release_object( fd );
1833 /* cancels all async I/O */
1834 DECL_HANDLER(cancel_async)
1836 struct fd *fd = get_handle_fd_obj( current->process, req->handle, 0 );
1839 /* Note: we don't kill the queued APC_ASYNC_IO on this thread because
1840 * NtCancelIoFile() will force the pending APC to be run. Since,
1841 * Windows only guarantees that the current thread will have no async
1842 * operation on the current fd when NtCancelIoFile returns, this shall
1845 fd->fd_ops->cancel_async( fd );
1846 release_object( fd );