4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
128 #include <linux/vs_limit.h>
130 #include <asm/semaphore.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable = 1;
138 int lease_break_time = 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 static LIST_HEAD(file_lock_list);
144 static LIST_HEAD(blocked_list);
146 static kmem_cache_t *filelock_cache __read_mostly;
148 /* Allocate an empty lock structure. */
149 static struct file_lock *locks_alloc_lock(void)
151 if (!vx_locks_avail(1))
153 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
156 static void locks_release_private(struct file_lock *fl)
159 if (fl->fl_ops->fl_release_private)
160 fl->fl_ops->fl_release_private(fl);
164 if (fl->fl_lmops->fl_release_private)
165 fl->fl_lmops->fl_release_private(fl);
171 /* Free a lock which is not in use. */
172 static void locks_free_lock(struct file_lock *fl)
174 BUG_ON(waitqueue_active(&fl->fl_wait));
175 BUG_ON(!list_empty(&fl->fl_block));
176 BUG_ON(!list_empty(&fl->fl_link));
179 locks_release_private(fl);
180 kmem_cache_free(filelock_cache, fl);
183 void locks_init_lock(struct file_lock *fl)
185 INIT_LIST_HEAD(&fl->fl_link);
186 INIT_LIST_HEAD(&fl->fl_block);
187 init_waitqueue_head(&fl->fl_wait);
189 fl->fl_fasync = NULL;
195 fl->fl_start = fl->fl_end = 0;
201 EXPORT_SYMBOL(locks_init_lock);
204 * Initialises the fields of the file lock which are invariant for
207 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
209 struct file_lock *lock = (struct file_lock *) foo;
211 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
212 SLAB_CTOR_CONSTRUCTOR)
215 locks_init_lock(lock);
218 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
221 if (fl->fl_ops->fl_copy_lock)
222 fl->fl_ops->fl_copy_lock(new, fl);
223 new->fl_ops = fl->fl_ops;
226 if (fl->fl_lmops->fl_copy_lock)
227 fl->fl_lmops->fl_copy_lock(new, fl);
228 new->fl_lmops = fl->fl_lmops;
233 * Initialize a new lock from an existing file_lock structure.
235 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
237 new->fl_owner = fl->fl_owner;
238 new->fl_pid = fl->fl_pid;
240 new->fl_flags = fl->fl_flags;
241 new->fl_type = fl->fl_type;
242 new->fl_start = fl->fl_start;
243 new->fl_end = fl->fl_end;
245 new->fl_lmops = NULL;
248 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
250 locks_release_private(new);
252 __locks_copy_lock(new, fl);
253 new->fl_file = fl->fl_file;
254 new->fl_ops = fl->fl_ops;
255 new->fl_lmops = fl->fl_lmops;
256 new->fl_xid = fl->fl_xid;
258 locks_copy_private(new, fl);
261 EXPORT_SYMBOL(locks_copy_lock);
263 static inline int flock_translate_cmd(int cmd) {
265 return cmd & (LOCK_MAND | LOCK_RW);
277 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
278 static int flock_make_lock(struct file *filp, struct file_lock **lock,
281 struct file_lock *fl;
282 int type = flock_translate_cmd(cmd);
286 fl = locks_alloc_lock();
291 fl->fl_pid = current->tgid;
292 fl->fl_flags = FL_FLOCK;
294 fl->fl_end = OFFSET_MAX;
296 vxd_assert(filp->f_xid == vx_current_xid(),
297 "f_xid(%d) == current(%d)", filp->f_xid, vx_current_xid());
298 fl->fl_xid = filp->f_xid;
305 static int assign_type(struct file_lock *fl, int type)
319 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
322 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
327 switch (l->l_whence) {
335 start = i_size_read(filp->f_dentry->d_inode);
341 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
342 POSIX-2001 defines it. */
346 fl->fl_end = OFFSET_MAX;
348 end = start + l->l_len - 1;
350 } else if (l->l_len < 0) {
357 fl->fl_start = start; /* we record the absolute position */
358 if (fl->fl_end < fl->fl_start)
361 fl->fl_owner = current->files;
362 fl->fl_pid = current->tgid;
364 fl->fl_flags = FL_POSIX;
368 return assign_type(fl, l->l_type);
371 #if BITS_PER_LONG == 32
372 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
377 switch (l->l_whence) {
385 start = i_size_read(filp->f_dentry->d_inode);
394 fl->fl_end = OFFSET_MAX;
396 fl->fl_end = start + l->l_len - 1;
397 } else if (l->l_len < 0) {
398 fl->fl_end = start - 1;
403 fl->fl_start = start; /* we record the absolute position */
404 if (fl->fl_end < fl->fl_start)
407 fl->fl_owner = current->files;
408 fl->fl_pid = current->tgid;
410 fl->fl_flags = FL_POSIX;
418 fl->fl_type = l->l_type;
428 /* default lease lock manager operations */
429 static void lease_break_callback(struct file_lock *fl)
431 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
434 static void lease_release_private_callback(struct file_lock *fl)
439 f_delown(fl->fl_file);
440 fl->fl_file->f_owner.signum = 0;
443 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
445 return fl->fl_file == try->fl_file;
448 static struct lock_manager_operations lease_manager_ops = {
449 .fl_break = lease_break_callback,
450 .fl_release_private = lease_release_private_callback,
451 .fl_mylease = lease_mylease_callback,
452 .fl_change = lease_modify,
456 * Initialize a lease, use the default lock manager operations
458 static int lease_init(struct file *filp, int type, struct file_lock *fl)
460 if (assign_type(fl, type) != 0)
463 fl->fl_owner = current->files;
464 fl->fl_pid = current->tgid;
465 fl->fl_xid = vx_current_xid();
468 fl->fl_flags = FL_LEASE;
470 fl->fl_end = OFFSET_MAX;
472 fl->fl_lmops = &lease_manager_ops;
476 /* Allocate a file_lock initialised to this type of lease */
477 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
479 struct file_lock *fl = locks_alloc_lock();
485 fl->fl_xid = vx_current_xid();
487 vxd_assert(filp->f_xid == fl->fl_xid,
488 "f_xid(%d) == fl_xid(%d)", filp->f_xid, fl->fl_xid);
490 error = lease_init(filp, type, fl);
500 /* Check if two locks overlap each other.
502 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
504 return ((fl1->fl_end >= fl2->fl_start) &&
505 (fl2->fl_end >= fl1->fl_start));
509 * Check whether two locks have the same owner.
511 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
513 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
514 return fl2->fl_lmops == fl1->fl_lmops &&
515 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
516 return fl1->fl_owner == fl2->fl_owner;
519 /* Remove waiter from blocker's block list.
520 * When blocker ends up pointing to itself then the list is empty.
522 static void __locks_delete_block(struct file_lock *waiter)
524 list_del_init(&waiter->fl_block);
525 list_del_init(&waiter->fl_link);
526 waiter->fl_next = NULL;
531 static void locks_delete_block(struct file_lock *waiter)
534 __locks_delete_block(waiter);
538 /* Insert waiter into blocker's block list.
539 * We use a circular list so that processes can be easily woken up in
540 * the order they blocked. The documentation doesn't require this but
541 * it seems like the reasonable thing to do.
543 static void locks_insert_block(struct file_lock *blocker,
544 struct file_lock *waiter)
546 BUG_ON(!list_empty(&waiter->fl_block));
547 list_add_tail(&waiter->fl_block, &blocker->fl_block);
548 waiter->fl_next = blocker;
549 if (IS_POSIX(blocker))
550 list_add(&waiter->fl_link, &blocked_list);
553 /* Wake up processes blocked waiting for blocker.
554 * If told to wait then schedule the processes until the block list
555 * is empty, otherwise empty the block list ourselves.
557 static void locks_wake_up_blocks(struct file_lock *blocker)
559 while (!list_empty(&blocker->fl_block)) {
560 struct file_lock *waiter = list_entry(blocker->fl_block.next,
561 struct file_lock, fl_block);
562 __locks_delete_block(waiter);
563 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
564 waiter->fl_lmops->fl_notify(waiter);
566 wake_up(&waiter->fl_wait);
570 /* Insert file lock fl into an inode's lock list at the position indicated
571 * by pos. At the same time add the lock to the global file lock list.
573 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
575 list_add(&fl->fl_link, &file_lock_list);
577 /* insert into file's list */
581 if (fl->fl_ops && fl->fl_ops->fl_insert)
582 fl->fl_ops->fl_insert(fl);
586 * Delete a lock and then free it.
587 * Wake up processes that are blocked waiting for this lock,
588 * notify the FS that the lock has been cleared and
589 * finally free the lock.
591 static void locks_delete_lock(struct file_lock **thisfl_p)
593 struct file_lock *fl = *thisfl_p;
595 *thisfl_p = fl->fl_next;
597 list_del_init(&fl->fl_link);
599 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
600 if (fl->fl_fasync != NULL) {
601 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
602 fl->fl_fasync = NULL;
605 if (fl->fl_ops && fl->fl_ops->fl_remove)
606 fl->fl_ops->fl_remove(fl);
608 locks_wake_up_blocks(fl);
612 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
613 * checks for shared/exclusive status of overlapping locks.
615 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
617 if (sys_fl->fl_type == F_WRLCK)
619 if (caller_fl->fl_type == F_WRLCK)
624 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
625 * checking before calling the locks_conflict().
627 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
629 /* POSIX locks owned by the same process do not conflict with
632 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
635 /* Check whether they overlap */
636 if (!locks_overlap(caller_fl, sys_fl))
639 return (locks_conflict(caller_fl, sys_fl));
642 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
643 * checking before calling the locks_conflict().
645 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
647 /* FLOCK locks referring to the same filp do not conflict with
650 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
652 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
655 return (locks_conflict(caller_fl, sys_fl));
658 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
661 DECLARE_WAITQUEUE(wait, current);
663 __set_current_state(TASK_INTERRUPTIBLE);
664 add_wait_queue(fl_wait, &wait);
668 result = schedule_timeout(timeout);
669 if (signal_pending(current))
670 result = -ERESTARTSYS;
671 remove_wait_queue(fl_wait, &wait);
672 __set_current_state(TASK_RUNNING);
676 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
679 locks_insert_block(blocker, waiter);
680 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
681 __locks_delete_block(waiter);
686 posix_test_lock(struct file *filp, struct file_lock *fl,
687 struct file_lock *conflock)
689 struct file_lock *cfl;
692 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
695 if (posix_locks_conflict(cfl, fl))
699 __locks_copy_lock(conflock, cfl);
707 EXPORT_SYMBOL(posix_test_lock);
709 /* This function tests for deadlock condition before putting a process to
710 * sleep. The detection scheme is no longer recursive. Recursive was neat,
711 * but dangerous - we risked stack corruption if the lock data was bad, or
712 * if the recursion was too deep for any other reason.
714 * We rely on the fact that a task can only be on one lock's wait queue
715 * at a time. When we find blocked_task on a wait queue we can re-search
716 * with blocked_task equal to that queue's owner, until either blocked_task
717 * isn't found, or blocked_task is found on a queue owned by my_task.
719 * Note: the above assumption may not be true when handling lock requests
720 * from a broken NFS client. But broken NFS clients have a lot more to
721 * worry about than proper deadlock detection anyway... --okir
723 int posix_locks_deadlock(struct file_lock *caller_fl,
724 struct file_lock *block_fl)
726 struct list_head *tmp;
729 if (posix_same_owner(caller_fl, block_fl))
731 list_for_each(tmp, &blocked_list) {
732 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
733 if (posix_same_owner(fl, block_fl)) {
742 EXPORT_SYMBOL(posix_locks_deadlock);
744 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
745 * at the head of the list, but that's secret knowledge known only to
746 * flock_lock_file and posix_lock_file.
748 static int flock_lock_file(struct file *filp, struct file_lock *request)
750 struct file_lock *new_fl = NULL;
751 struct file_lock **before;
752 struct inode * inode = filp->f_dentry->d_inode;
757 for_each_lock(inode, before) {
758 struct file_lock *fl = *before;
763 if (filp != fl->fl_file)
765 if (request->fl_type == fl->fl_type)
768 locks_delete_lock(before);
772 if (request->fl_type == F_UNLCK)
776 new_fl = locks_alloc_lock();
780 * If a higher-priority process was blocked on the old file lock,
781 * give it the opportunity to lock the file.
786 for_each_lock(inode, before) {
787 struct file_lock *fl = *before;
792 if (!flock_locks_conflict(request, fl))
795 if (request->fl_flags & FL_SLEEP)
796 locks_insert_block(fl, request);
799 locks_copy_lock(new_fl, request);
800 vx_locks_inc(new_fl);
801 locks_insert_lock(&inode->i_flock, new_fl);
808 locks_free_lock(new_fl);
812 static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request,
813 struct file_lock *conflock, xid_t xid)
815 struct file_lock *fl;
816 struct file_lock *new_fl, *new_fl2;
817 struct file_lock *left = NULL;
818 struct file_lock *right = NULL;
819 struct file_lock **before;
820 int error, added = 0;
822 vxd_assert(xid == vx_current_xid(),
823 "xid(%d) == current(%d)", xid, vx_current_xid());
825 * We may need two file_lock structures for this operation,
826 * so we get them in advance to avoid races.
828 new_fl = locks_alloc_lock();
829 new_fl->fl_xid = xid;
830 vx_locks_inc(new_fl);
831 new_fl2 = locks_alloc_lock();
832 new_fl2->fl_xid = xid;
833 vx_locks_inc(new_fl2);
836 if (request->fl_type != F_UNLCK) {
837 for_each_lock(inode, before) {
838 struct file_lock *fl = *before;
841 if (!posix_locks_conflict(request, fl))
844 locks_copy_lock(conflock, fl);
846 if (!(request->fl_flags & FL_SLEEP))
849 if (posix_locks_deadlock(request, fl))
852 locks_insert_block(fl, request);
857 /* If we're just looking for a conflict, we're done. */
859 if (request->fl_flags & FL_ACCESS)
862 error = -ENOLCK; /* "no luck" */
863 if (!(new_fl && new_fl2))
867 * We've allocated the new locks in advance, so there are no
868 * errors possible (and no blocking operations) from here on.
870 * Find the first old lock with the same owner as the new lock.
873 before = &inode->i_flock;
875 /* First skip locks owned by other processes. */
876 while ((fl = *before) && (!IS_POSIX(fl) ||
877 !posix_same_owner(request, fl))) {
878 before = &fl->fl_next;
881 /* Process locks with this owner. */
882 while ((fl = *before) && posix_same_owner(request, fl)) {
883 /* Detect adjacent or overlapping regions (if same lock type)
885 if (request->fl_type == fl->fl_type) {
886 /* In all comparisons of start vs end, use
887 * "start - 1" rather than "end + 1". If end
888 * is OFFSET_MAX, end + 1 will become negative.
890 if (fl->fl_end < request->fl_start - 1)
892 /* If the next lock in the list has entirely bigger
893 * addresses than the new one, insert the lock here.
895 if (fl->fl_start - 1 > request->fl_end)
898 /* If we come here, the new and old lock are of the
899 * same type and adjacent or overlapping. Make one
900 * lock yielding from the lower start address of both
901 * locks to the higher end address.
903 if (fl->fl_start > request->fl_start)
904 fl->fl_start = request->fl_start;
906 request->fl_start = fl->fl_start;
907 if (fl->fl_end < request->fl_end)
908 fl->fl_end = request->fl_end;
910 request->fl_end = fl->fl_end;
912 locks_delete_lock(before);
919 /* Processing for different lock types is a bit
922 if (fl->fl_end < request->fl_start)
924 if (fl->fl_start > request->fl_end)
926 if (request->fl_type == F_UNLCK)
928 if (fl->fl_start < request->fl_start)
930 /* If the next lock in the list has a higher end
931 * address than the new one, insert the new one here.
933 if (fl->fl_end > request->fl_end) {
937 if (fl->fl_start >= request->fl_start) {
938 /* The new lock completely replaces an old
939 * one (This may happen several times).
942 locks_delete_lock(before);
945 /* Replace the old lock with the new one.
946 * Wake up anybody waiting for the old one,
947 * as the change in lock type might satisfy
950 locks_wake_up_blocks(fl);
951 fl->fl_start = request->fl_start;
952 fl->fl_end = request->fl_end;
953 fl->fl_type = request->fl_type;
954 locks_release_private(fl);
955 locks_copy_private(fl, request);
960 /* Go on to next lock.
963 before = &fl->fl_next;
968 if (request->fl_type == F_UNLCK)
970 locks_copy_lock(new_fl, request);
971 locks_insert_lock(before, new_fl);
976 /* The new lock breaks the old one in two pieces,
977 * so we have to use the second new lock.
981 locks_copy_lock(left, right);
982 locks_insert_lock(before, left);
984 right->fl_start = request->fl_end + 1;
985 locks_wake_up_blocks(right);
988 left->fl_end = request->fl_start - 1;
989 locks_wake_up_blocks(left);
994 * Free any unused locks.
997 locks_free_lock(new_fl);
999 locks_free_lock(new_fl2);
1004 * posix_lock_file - Apply a POSIX-style lock to a file
1005 * @filp: The file to apply the lock to
1006 * @fl: The lock to be applied
1008 * Add a POSIX style lock to a file.
1009 * We merge adjacent & overlapping locks whenever possible.
1010 * POSIX locks are sorted by owner task, then by starting address
1012 int posix_lock_file(struct file *filp, struct file_lock *fl)
1014 return __posix_lock_file_conf(filp->f_dentry->d_inode,
1015 fl, NULL, filp->f_xid);
1017 EXPORT_SYMBOL(posix_lock_file);
1020 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1021 * @filp: The file to apply the lock to
1022 * @fl: The lock to be applied
1023 * @conflock: Place to return a copy of the conflicting lock, if found.
1025 * Except for the conflock parameter, acts just like posix_lock_file.
1027 int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1028 struct file_lock *conflock)
1030 return __posix_lock_file_conf(filp->f_dentry->d_inode,
1031 fl, conflock, filp->f_xid);
1033 EXPORT_SYMBOL(posix_lock_file_conf);
1036 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1037 * @filp: The file to apply the lock to
1038 * @fl: The lock to be applied
1040 * Add a POSIX style lock to a file.
1041 * We merge adjacent & overlapping locks whenever possible.
1042 * POSIX locks are sorted by owner task, then by starting address
1044 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1049 error = posix_lock_file(filp, fl);
1050 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1052 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1056 locks_delete_block(fl);
1061 EXPORT_SYMBOL(posix_lock_file_wait);
1064 * locks_mandatory_locked - Check for an active lock
1065 * @inode: the file to check
1067 * Searches the inode's list of locks to find any POSIX locks which conflict.
1068 * This function is called from locks_verify_locked() only.
1070 int locks_mandatory_locked(struct inode *inode)
1072 fl_owner_t owner = current->files;
1073 struct file_lock *fl;
1076 * Search the lock list for this inode for any POSIX locks.
1079 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1082 if (fl->fl_owner != owner)
1086 return fl ? -EAGAIN : 0;
1090 * locks_mandatory_area - Check for a conflicting lock
1091 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1093 * @inode: the file to check
1094 * @filp: how the file was opened (if it was)
1095 * @offset: start of area to check
1096 * @count: length of area to check
1098 * Searches the inode's list of locks to find any POSIX locks which conflict.
1099 * This function is called from rw_verify_area() and
1100 * locks_verify_truncate().
1102 int locks_mandatory_area(int read_write, struct inode *inode,
1103 struct file *filp, loff_t offset,
1106 struct file_lock fl;
1109 locks_init_lock(&fl);
1110 fl.fl_owner = current->files;
1111 fl.fl_pid = current->tgid;
1113 fl.fl_flags = FL_POSIX | FL_ACCESS;
1114 if (filp && !(filp->f_flags & O_NONBLOCK))
1115 fl.fl_flags |= FL_SLEEP;
1116 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1117 fl.fl_start = offset;
1118 fl.fl_end = offset + count - 1;
1121 error = __posix_lock_file_conf(inode, &fl, NULL, filp->f_xid);
1122 if (error != -EAGAIN)
1124 if (!(fl.fl_flags & FL_SLEEP))
1126 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1129 * If we've been sleeping someone might have
1130 * changed the permissions behind our back.
1132 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1136 locks_delete_block(&fl);
1143 EXPORT_SYMBOL(locks_mandatory_area);
1145 /* We already had a lease on this file; just change its type */
1146 int lease_modify(struct file_lock **before, int arg)
1148 struct file_lock *fl = *before;
1149 int error = assign_type(fl, arg);
1153 locks_wake_up_blocks(fl);
1155 locks_delete_lock(before);
1159 EXPORT_SYMBOL(lease_modify);
1161 static void time_out_leases(struct inode *inode)
1163 struct file_lock **before;
1164 struct file_lock *fl;
1166 before = &inode->i_flock;
1167 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1168 if ((fl->fl_break_time == 0)
1169 || time_before(jiffies, fl->fl_break_time)) {
1170 before = &fl->fl_next;
1173 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1174 if (fl == *before) /* lease_modify may have freed fl */
1175 before = &fl->fl_next;
1180 * __break_lease - revoke all outstanding leases on file
1181 * @inode: the inode of the file to return
1182 * @mode: the open mode (read or write)
1184 * break_lease (inlined for speed) has checked there already
1185 * is a lease on this file. Leases are broken on a call to open()
1186 * or truncate(). This function can sleep unless you
1187 * specified %O_NONBLOCK to your open().
1189 int __break_lease(struct inode *inode, unsigned int mode)
1191 int error = 0, future;
1192 struct file_lock *new_fl, *flock;
1193 struct file_lock *fl;
1195 unsigned long break_time;
1196 int i_have_this_lease = 0;
1198 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1203 time_out_leases(inode);
1205 flock = inode->i_flock;
1206 if ((flock == NULL) || !IS_LEASE(flock))
1209 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1210 if (fl->fl_owner == current->files)
1211 i_have_this_lease = 1;
1213 if (mode & FMODE_WRITE) {
1214 /* If we want write access, we have to revoke any lease. */
1215 future = F_UNLCK | F_INPROGRESS;
1216 } else if (flock->fl_type & F_INPROGRESS) {
1217 /* If the lease is already being broken, we just leave it */
1218 future = flock->fl_type;
1219 } else if (flock->fl_type & F_WRLCK) {
1220 /* Downgrade the exclusive lease to a read-only lease. */
1221 future = F_RDLCK | F_INPROGRESS;
1223 /* the existing lease was read-only, so we can read too. */
1227 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1233 if (lease_break_time > 0) {
1234 break_time = jiffies + lease_break_time * HZ;
1235 if (break_time == 0)
1236 break_time++; /* so that 0 means no break time */
1239 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1240 if (fl->fl_type != future) {
1241 fl->fl_type = future;
1242 fl->fl_break_time = break_time;
1243 /* lease must have lmops break callback */
1244 fl->fl_lmops->fl_break(fl);
1248 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1249 error = -EWOULDBLOCK;
1254 break_time = flock->fl_break_time;
1255 if (break_time != 0) {
1256 break_time -= jiffies;
1257 if (break_time == 0)
1260 error = locks_block_on_timeout(flock, new_fl, break_time);
1263 time_out_leases(inode);
1264 /* Wait for the next lease that has not been broken yet */
1265 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1266 flock = flock->fl_next) {
1267 if (flock->fl_type & F_INPROGRESS)
1276 locks_free_lock(new_fl);
1280 EXPORT_SYMBOL(__break_lease);
1285 * @time: pointer to a timespec which will contain the last modified time
1287 * This is to force NFS clients to flush their caches for files with
1288 * exclusive leases. The justification is that if someone has an
1289 * exclusive lease, then they could be modifiying it.
1291 void lease_get_mtime(struct inode *inode, struct timespec *time)
1293 struct file_lock *flock = inode->i_flock;
1294 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1295 *time = current_fs_time(inode->i_sb);
1297 *time = inode->i_mtime;
1300 EXPORT_SYMBOL(lease_get_mtime);
1303 * fcntl_getlease - Enquire what lease is currently active
1306 * The value returned by this function will be one of
1307 * (if no lease break is pending):
1309 * %F_RDLCK to indicate a shared lease is held.
1311 * %F_WRLCK to indicate an exclusive lease is held.
1313 * %F_UNLCK to indicate no lease is held.
1315 * (if a lease break is pending):
1317 * %F_RDLCK to indicate an exclusive lease needs to be
1318 * changed to a shared lease (or removed).
1320 * %F_UNLCK to indicate the lease needs to be removed.
1322 * XXX: sfr & willy disagree over whether F_INPROGRESS
1323 * should be returned to userspace.
1325 int fcntl_getlease(struct file *filp)
1327 struct file_lock *fl;
1331 time_out_leases(filp->f_dentry->d_inode);
1332 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1334 if (fl->fl_file == filp) {
1335 type = fl->fl_type & ~F_INPROGRESS;
1344 * __setlease - sets a lease on an open file
1345 * @filp: file pointer
1346 * @arg: type of lease to obtain
1347 * @flp: input - file_lock to use, output - file_lock inserted
1349 * The (input) flp->fl_lmops->fl_break function is required
1352 * Called with kernel lock held.
1354 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1356 struct file_lock *fl, **before, **my_before = NULL, *lease;
1357 struct dentry *dentry = filp->f_dentry;
1358 struct inode *inode = dentry->d_inode;
1359 int error, rdlease_count = 0, wrlease_count = 0;
1361 time_out_leases(inode);
1364 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1370 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1372 if ((arg == F_WRLCK)
1373 && ((atomic_read(&dentry->d_count) > 1)
1374 || (atomic_read(&inode->i_count) > 1)))
1378 * At this point, we know that if there is an exclusive
1379 * lease on this file, then we hold it on this filp
1380 * (otherwise our open of this file would have blocked).
1381 * And if we are trying to acquire an exclusive lease,
1382 * then the file is not open by anyone (including us)
1383 * except for this filp.
1385 for (before = &inode->i_flock;
1386 ((fl = *before) != NULL) && IS_LEASE(fl);
1387 before = &fl->fl_next) {
1388 if (lease->fl_lmops->fl_mylease(fl, lease))
1390 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1392 * Someone is in the process of opening this
1393 * file for writing so we may not take an
1394 * exclusive lease on it.
1401 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1402 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1405 if (my_before != NULL) {
1407 error = lease->fl_lmops->fl_change(my_before, arg);
1419 error = lease_alloc(filp, arg, &fl);
1423 locks_copy_lock(fl, lease);
1425 locks_insert_lock(before, fl);
1433 * setlease - sets a lease on an open file
1434 * @filp: file pointer
1435 * @arg: type of lease to obtain
1436 * @lease: file_lock to use
1438 * Call this to establish a lease on the file.
1439 * The fl_lmops fl_break function is required by break_lease
1442 int setlease(struct file *filp, long arg, struct file_lock **lease)
1444 struct dentry *dentry = filp->f_dentry;
1445 struct inode *inode = dentry->d_inode;
1448 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1450 if (!S_ISREG(inode->i_mode))
1452 error = security_file_lock(filp, arg);
1457 error = __setlease(filp, arg, lease);
1463 EXPORT_SYMBOL(setlease);
1466 * fcntl_setlease - sets a lease on an open file
1467 * @fd: open file descriptor
1468 * @filp: file pointer
1469 * @arg: type of lease to obtain
1471 * Call this fcntl to establish a lease on the file.
1472 * Note that you also need to call %F_SETSIG to
1473 * receive a signal when the lease is broken.
1475 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1477 struct file_lock fl, *flp = &fl;
1478 struct dentry *dentry = filp->f_dentry;
1479 struct inode *inode = dentry->d_inode;
1482 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1484 if (!S_ISREG(inode->i_mode))
1486 error = security_file_lock(filp, arg);
1490 locks_init_lock(&fl);
1491 error = lease_init(filp, arg, &fl);
1497 error = __setlease(filp, arg, &flp);
1498 if (error || arg == F_UNLCK)
1501 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1503 /* remove lease just inserted by __setlease */
1504 flp->fl_type = F_UNLCK | F_INPROGRESS;
1505 flp->fl_break_time = jiffies- 10;
1506 time_out_leases(inode);
1510 error = f_setown(filp, current->pid, 0);
1517 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1518 * @filp: The file to apply the lock to
1519 * @fl: The lock to be applied
1521 * Add a FLOCK style lock to a file.
1523 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1528 error = flock_lock_file(filp, fl);
1529 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1531 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1535 locks_delete_block(fl);
1541 EXPORT_SYMBOL(flock_lock_file_wait);
1544 * sys_flock: - flock() system call.
1545 * @fd: the file descriptor to lock.
1546 * @cmd: the type of lock to apply.
1548 * Apply a %FL_FLOCK style lock to an open file descriptor.
1549 * The @cmd can be one of
1551 * %LOCK_SH -- a shared lock.
1553 * %LOCK_EX -- an exclusive lock.
1555 * %LOCK_UN -- remove an existing lock.
1557 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1559 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1560 * processes read and write access respectively.
1562 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1565 struct file_lock *lock;
1566 int can_sleep, unlock;
1574 can_sleep = !(cmd & LOCK_NB);
1576 unlock = (cmd == LOCK_UN);
1578 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1581 error = flock_make_lock(filp, &lock, cmd);
1585 lock->fl_flags |= FL_SLEEP;
1587 error = security_file_lock(filp, cmd);
1591 if (filp->f_op && filp->f_op->flock)
1592 error = filp->f_op->flock(filp,
1593 (can_sleep) ? F_SETLKW : F_SETLK,
1596 error = flock_lock_file_wait(filp, lock);
1599 locks_free_lock(lock);
1607 /* Report the first existing lock that would conflict with l.
1608 * This implements the F_GETLK command of fcntl().
1610 int fcntl_getlk(struct file *filp, struct flock __user *l)
1612 struct file_lock *fl, cfl, file_lock;
1617 if (copy_from_user(&flock, l, sizeof(flock)))
1620 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1623 error = flock_to_posix_lock(filp, &file_lock, &flock);
1627 if (filp->f_op && filp->f_op->lock) {
1628 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1629 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1630 file_lock.fl_ops->fl_release_private(&file_lock);
1634 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1636 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1639 flock.l_type = F_UNLCK;
1641 flock.l_pid = fl->fl_pid;
1642 #if BITS_PER_LONG == 32
1644 * Make sure we can represent the posix lock via
1645 * legacy 32bit flock.
1648 if (fl->fl_start > OFFT_OFFSET_MAX)
1650 if ((fl->fl_end != OFFSET_MAX)
1651 && (fl->fl_end > OFFT_OFFSET_MAX))
1654 flock.l_start = fl->fl_start;
1655 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1656 fl->fl_end - fl->fl_start + 1;
1658 flock.l_type = fl->fl_type;
1661 if (!copy_to_user(l, &flock, sizeof(flock)))
1667 /* Apply the lock described by l to an open file descriptor.
1668 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1670 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1671 struct flock __user *l)
1673 struct file_lock *file_lock = locks_alloc_lock();
1675 struct inode *inode;
1678 if (file_lock == NULL)
1681 vxd_assert(filp->f_xid == vx_current_xid(),
1682 "f_xid(%d) == current(%d)", filp->f_xid, vx_current_xid());
1683 file_lock->fl_xid = filp->f_xid;
1684 vx_locks_inc(file_lock);
1687 * This might block, so we do it before checking the inode.
1690 if (copy_from_user(&flock, l, sizeof(flock)))
1693 inode = filp->f_dentry->d_inode;
1695 /* Don't allow mandatory locks on files that may be memory mapped
1698 if (IS_MANDLOCK(inode) &&
1699 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1700 mapping_writably_mapped(filp->f_mapping)) {
1706 error = flock_to_posix_lock(filp, file_lock, &flock);
1709 if (cmd == F_SETLKW) {
1710 file_lock->fl_flags |= FL_SLEEP;
1714 switch (flock.l_type) {
1716 if (!(filp->f_mode & FMODE_READ))
1720 if (!(filp->f_mode & FMODE_WRITE))
1730 error = security_file_lock(filp, file_lock->fl_type);
1734 if (filp->f_op && filp->f_op->lock != NULL)
1735 error = filp->f_op->lock(filp, cmd, file_lock);
1738 error = posix_lock_file(filp, file_lock);
1739 if ((error != -EAGAIN) || (cmd == F_SETLK))
1741 error = wait_event_interruptible(file_lock->fl_wait,
1742 !file_lock->fl_next);
1746 locks_delete_block(file_lock);
1752 * Attempt to detect a close/fcntl race and recover by
1753 * releasing the lock that was just acquired.
1755 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1756 flock.l_type = F_UNLCK;
1761 locks_free_lock(file_lock);
1765 #if BITS_PER_LONG == 32
1766 /* Report the first existing lock that would conflict with l.
1767 * This implements the F_GETLK command of fcntl().
1769 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1771 struct file_lock *fl, cfl, file_lock;
1772 struct flock64 flock;
1776 if (copy_from_user(&flock, l, sizeof(flock)))
1779 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1782 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1786 if (filp->f_op && filp->f_op->lock) {
1787 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1788 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1789 file_lock.fl_ops->fl_release_private(&file_lock);
1793 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1795 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1798 flock.l_type = F_UNLCK;
1800 flock.l_pid = fl->fl_pid;
1801 flock.l_start = fl->fl_start;
1802 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1803 fl->fl_end - fl->fl_start + 1;
1805 flock.l_type = fl->fl_type;
1808 if (!copy_to_user(l, &flock, sizeof(flock)))
1815 /* Apply the lock described by l to an open file descriptor.
1816 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1818 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1819 struct flock64 __user *l)
1821 struct file_lock *file_lock = locks_alloc_lock();
1822 struct flock64 flock;
1823 struct inode *inode;
1826 if (file_lock == NULL)
1829 vxd_assert(filp->f_xid == vx_current_xid(),
1830 "f_xid(%d) == current(%d)", filp->f_xid, vx_current_xid());
1831 file_lock->fl_xid = filp->f_xid;
1832 vx_locks_inc(file_lock);
1835 * This might block, so we do it before checking the inode.
1838 if (copy_from_user(&flock, l, sizeof(flock)))
1841 inode = filp->f_dentry->d_inode;
1843 /* Don't allow mandatory locks on files that may be memory mapped
1846 if (IS_MANDLOCK(inode) &&
1847 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1848 mapping_writably_mapped(filp->f_mapping)) {
1854 error = flock64_to_posix_lock(filp, file_lock, &flock);
1857 if (cmd == F_SETLKW64) {
1858 file_lock->fl_flags |= FL_SLEEP;
1862 switch (flock.l_type) {
1864 if (!(filp->f_mode & FMODE_READ))
1868 if (!(filp->f_mode & FMODE_WRITE))
1878 error = security_file_lock(filp, file_lock->fl_type);
1882 if (filp->f_op && filp->f_op->lock != NULL)
1883 error = filp->f_op->lock(filp, cmd, file_lock);
1886 error = posix_lock_file(filp, file_lock);
1887 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1889 error = wait_event_interruptible(file_lock->fl_wait,
1890 !file_lock->fl_next);
1894 locks_delete_block(file_lock);
1900 * Attempt to detect a close/fcntl race and recover by
1901 * releasing the lock that was just acquired.
1903 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1904 flock.l_type = F_UNLCK;
1909 locks_free_lock(file_lock);
1912 #endif /* BITS_PER_LONG == 32 */
1915 * This function is called when the file is being removed
1916 * from the task's fd array. POSIX locks belonging to this task
1917 * are deleted at this time.
1919 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1921 struct file_lock lock, **before;
1924 * If there are no locks held on this file, we don't need to call
1925 * posix_lock_file(). Another process could be setting a lock on this
1926 * file at the same time, but we wouldn't remove that lock anyway.
1928 before = &filp->f_dentry->d_inode->i_flock;
1929 if (*before == NULL)
1932 lock.fl_type = F_UNLCK;
1933 lock.fl_flags = FL_POSIX;
1935 lock.fl_end = OFFSET_MAX;
1936 lock.fl_owner = owner;
1937 lock.fl_pid = current->tgid;
1938 lock.fl_file = filp;
1940 lock.fl_lmops = NULL;
1942 if (filp->f_op && filp->f_op->lock != NULL) {
1943 filp->f_op->lock(filp, F_SETLK, &lock);
1947 /* Can't use posix_lock_file here; we need to remove it no matter
1948 * which pid we have.
1951 while (*before != NULL) {
1952 struct file_lock *fl = *before;
1953 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1954 locks_delete_lock(before);
1957 before = &fl->fl_next;
1961 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1962 lock.fl_ops->fl_release_private(&lock);
1965 EXPORT_SYMBOL(locks_remove_posix);
1968 * This function is called on the last close of an open file.
1970 void locks_remove_flock(struct file *filp)
1972 struct inode * inode = filp->f_dentry->d_inode;
1973 struct file_lock *fl;
1974 struct file_lock **before;
1976 if (!inode->i_flock)
1979 if (filp->f_op && filp->f_op->flock) {
1980 struct file_lock fl = {
1981 .fl_pid = current->tgid,
1983 .fl_flags = FL_FLOCK,
1985 .fl_end = OFFSET_MAX,
1987 filp->f_op->flock(filp, F_SETLKW, &fl);
1988 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1989 fl.fl_ops->fl_release_private(&fl);
1993 before = &inode->i_flock;
1995 while ((fl = *before) != NULL) {
1996 if (fl->fl_file == filp) {
1998 locks_delete_lock(before);
2002 lease_modify(before, F_UNLCK);
2008 before = &fl->fl_next;
2014 * posix_unblock_lock - stop waiting for a file lock
2015 * @filp: how the file was opened
2016 * @waiter: the lock which was waiting
2018 * lockd needs to block waiting for locks.
2021 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2026 if (waiter->fl_next)
2027 __locks_delete_block(waiter);
2034 EXPORT_SYMBOL(posix_unblock_lock);
2036 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2038 struct inode *inode = NULL;
2040 if (fl->fl_file != NULL)
2041 inode = fl->fl_file->f_dentry->d_inode;
2043 out += sprintf(out, "%d:%s ", id, pfx);
2045 out += sprintf(out, "%6s %s ",
2046 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2047 (inode == NULL) ? "*NOINODE*" :
2048 (IS_MANDLOCK(inode) &&
2049 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2050 "MANDATORY" : "ADVISORY ");
2051 } else if (IS_FLOCK(fl)) {
2052 if (fl->fl_type & LOCK_MAND) {
2053 out += sprintf(out, "FLOCK MSNFS ");
2055 out += sprintf(out, "FLOCK ADVISORY ");
2057 } else if (IS_LEASE(fl)) {
2058 out += sprintf(out, "LEASE ");
2059 if (fl->fl_type & F_INPROGRESS)
2060 out += sprintf(out, "BREAKING ");
2061 else if (fl->fl_file)
2062 out += sprintf(out, "ACTIVE ");
2064 out += sprintf(out, "BREAKER ");
2066 out += sprintf(out, "UNKNOWN UNKNOWN ");
2068 if (fl->fl_type & LOCK_MAND) {
2069 out += sprintf(out, "%s ",
2070 (fl->fl_type & LOCK_READ)
2071 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2072 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2074 out += sprintf(out, "%s ",
2075 (fl->fl_type & F_INPROGRESS)
2076 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2077 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2080 #ifdef WE_CAN_BREAK_LSLK_NOW
2081 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2082 inode->i_sb->s_id, inode->i_ino);
2084 /* userspace relies on this representation of dev_t ;-( */
2085 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2086 MAJOR(inode->i_sb->s_dev),
2087 MINOR(inode->i_sb->s_dev), inode->i_ino);
2090 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2093 if (fl->fl_end == OFFSET_MAX)
2094 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2096 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2099 out += sprintf(out, "0 EOF\n");
2103 static void move_lock_status(char **p, off_t* pos, off_t offset)
2107 if(*pos >= offset) {
2108 /* the complete line is valid */
2113 if(*pos+len > offset) {
2114 /* use the second part of the line */
2115 int i = offset-*pos;
2116 memmove(*p,*p+i,len-i);
2121 /* discard the complete line */
2126 * get_locks_status - reports lock usage in /proc/locks
2127 * @buffer: address in userspace to write into
2129 * @offset: how far we are through the buffer
2130 * @length: how much to read
2133 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2135 struct list_head *tmp;
2141 list_for_each(tmp, &file_lock_list) {
2142 struct list_head *btmp;
2143 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2145 if (!vx_check(fl->fl_xid, VX_IDENT|VX_WATCH))
2148 lock_get_status(q, fl, ++i, "");
2149 move_lock_status(&q, &pos, offset);
2151 if(pos >= offset+length)
2154 list_for_each(btmp, &fl->fl_block) {
2155 struct file_lock *bfl = list_entry(btmp,
2156 struct file_lock, fl_block);
2157 lock_get_status(q, bfl, i, " ->");
2158 move_lock_status(&q, &pos, offset);
2160 if(pos >= offset+length)
2167 if(q-buffer < length)
2173 * lock_may_read - checks that the region is free of locks
2174 * @inode: the inode that is being read
2175 * @start: the first byte to read
2176 * @len: the number of bytes to read
2178 * Emulates Windows locking requirements. Whole-file
2179 * mandatory locks (share modes) can prohibit a read and
2180 * byte-range POSIX locks can prohibit a read if they overlap.
2182 * N.B. this function is only ever called
2183 * from knfsd and ownership of locks is never checked.
2185 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2187 struct file_lock *fl;
2190 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2192 if (fl->fl_type == F_RDLCK)
2194 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2196 } else if (IS_FLOCK(fl)) {
2197 if (!(fl->fl_type & LOCK_MAND))
2199 if (fl->fl_type & LOCK_READ)
2210 EXPORT_SYMBOL(lock_may_read);
2213 * lock_may_write - checks that the region is free of locks
2214 * @inode: the inode that is being written
2215 * @start: the first byte to write
2216 * @len: the number of bytes to write
2218 * Emulates Windows locking requirements. Whole-file
2219 * mandatory locks (share modes) can prohibit a write and
2220 * byte-range POSIX locks can prohibit a write if they overlap.
2222 * N.B. this function is only ever called
2223 * from knfsd and ownership of locks is never checked.
2225 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2227 struct file_lock *fl;
2230 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2232 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2234 } else if (IS_FLOCK(fl)) {
2235 if (!(fl->fl_type & LOCK_MAND))
2237 if (fl->fl_type & LOCK_WRITE)
2248 EXPORT_SYMBOL(lock_may_write);
2250 static inline void __steal_locks(struct file *file, fl_owner_t from)
2252 struct inode *inode = file->f_dentry->d_inode;
2253 struct file_lock *fl = inode->i_flock;
2256 if (fl->fl_file == file && fl->fl_owner == from)
2257 fl->fl_owner = current->files;
2262 /* When getting ready for executing a binary, we make sure that current
2263 * has a files_struct on its own. Before dropping the old files_struct,
2264 * we take over ownership of all locks for all file descriptors we own.
2265 * Note that we may accidentally steal a lock for a file that a sibling
2266 * has created since the unshare_files() call.
2268 void steal_locks(fl_owner_t from)
2270 struct files_struct *files = current->files;
2272 struct fdtable *fdt;
2281 * We are not taking a ref to the file structures, so
2282 * we need to acquire ->file_lock.
2284 spin_lock(&files->file_lock);
2285 fdt = files_fdtable(files);
2289 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2291 set = fdt->open_fds->fds_bits[j++];
2294 struct file *file = fdt->fd[i];
2296 __steal_locks(file, from);
2302 spin_unlock(&files->file_lock);
2305 EXPORT_SYMBOL(steal_locks);
2307 static int __init filelock_init(void)
2309 filelock_cache = kmem_cache_create("file_lock_cache",
2310 sizeof(struct file_lock), 0, SLAB_PANIC,
2315 core_initcall(filelock_init);