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/vs_limit.h>
129 #include <asm/semaphore.h>
130 #include <asm/uaccess.h>
132 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136 int leases_enable = 1;
137 int lease_break_time = 45;
139 #define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142 LIST_HEAD(file_lock_list);
144 EXPORT_SYMBOL(file_lock_list);
146 static LIST_HEAD(blocked_list);
148 static kmem_cache_t *filelock_cache;
150 /* Allocate an empty lock structure. */
151 static struct file_lock *locks_alloc_lock(void)
153 if (!vx_locks_avail(1))
155 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
158 /* Free a lock which is not in use. */
159 static inline void locks_free_lock(struct file_lock *fl)
167 if (waitqueue_active(&fl->fl_wait))
168 panic("Attempting to free lock with active wait queue");
170 if (!list_empty(&fl->fl_block))
171 panic("Attempting to free lock with active block list");
173 if (!list_empty(&fl->fl_link))
174 panic("Attempting to free lock on active lock list");
177 if (fl->fl_ops->fl_release_private)
178 fl->fl_ops->fl_release_private(fl);
183 if (fl->fl_lmops->fl_release_private)
184 fl->fl_lmops->fl_release_private(fl);
188 kmem_cache_free(filelock_cache, fl);
191 void locks_init_lock(struct file_lock *fl)
193 INIT_LIST_HEAD(&fl->fl_link);
194 INIT_LIST_HEAD(&fl->fl_block);
195 init_waitqueue_head(&fl->fl_wait);
197 fl->fl_fasync = NULL;
203 fl->fl_start = fl->fl_end = 0;
209 EXPORT_SYMBOL(locks_init_lock);
212 * Initialises the fields of the file lock which are invariant for
215 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
217 struct file_lock *lock = (struct file_lock *) foo;
219 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
220 SLAB_CTOR_CONSTRUCTOR)
223 locks_init_lock(lock);
227 * Initialize a new lock from an existing file_lock structure.
229 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
231 new->fl_owner = fl->fl_owner;
232 new->fl_pid = fl->fl_pid;
233 new->fl_file = fl->fl_file;
234 new->fl_flags = fl->fl_flags;
235 new->fl_type = fl->fl_type;
236 new->fl_start = fl->fl_start;
237 new->fl_end = fl->fl_end;
238 new->fl_ops = fl->fl_ops;
239 new->fl_lmops = fl->fl_lmops;
240 if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
241 fl->fl_ops->fl_copy_lock(new, fl);
242 if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
243 fl->fl_lmops->fl_copy_lock(new, fl);
245 new->fl_xid = fl->fl_xid;
248 EXPORT_SYMBOL(locks_copy_lock);
250 static inline int flock_translate_cmd(int cmd) {
252 return cmd & (LOCK_MAND | LOCK_RW);
264 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
265 static int flock_make_lock(struct file *filp, struct file_lock **lock,
268 struct file_lock *fl;
269 int type = flock_translate_cmd(cmd);
273 fl = locks_alloc_lock();
278 fl->fl_pid = current->tgid;
279 fl->fl_flags = FL_FLOCK;
281 fl->fl_end = OFFSET_MAX;
283 /* check agains file xid maybe ? */
284 fl->fl_xid = vx_current_xid();
291 static int assign_type(struct file_lock *fl, int type)
305 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
308 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
313 switch (l->l_whence) {
321 start = i_size_read(filp->f_dentry->d_inode);
327 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
328 POSIX-2001 defines it. */
330 end = start + l->l_len - 1;
338 if (l->l_len > 0 && end < 0)
341 fl->fl_start = start; /* we record the absolute position */
344 fl->fl_end = OFFSET_MAX;
346 fl->fl_owner = current->files;
347 fl->fl_pid = current->tgid;
349 fl->fl_flags = FL_POSIX;
353 return assign_type(fl, l->l_type);
356 #if BITS_PER_LONG == 32
357 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 switch (l->l_whence) {
370 start = i_size_read(filp->f_dentry->d_inode);
376 if (((start += l->l_start) < 0) || (l->l_len < 0))
378 fl->fl_end = start + l->l_len - 1;
379 if (l->l_len > 0 && fl->fl_end < 0)
381 fl->fl_start = start; /* we record the absolute position */
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_owner = current->files;
386 fl->fl_pid = current->tgid;
388 fl->fl_flags = FL_POSIX;
396 fl->fl_type = l->l_type;
406 /* default lease lock manager operations */
407 static void lease_break_callback(struct file_lock *fl)
409 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
412 static void lease_release_private_callback(struct file_lock *fl)
417 f_delown(fl->fl_file);
418 fl->fl_file->f_owner.signum = 0;
421 struct lock_manager_operations lease_manager_ops = {
422 .fl_break = lease_break_callback,
423 .fl_release_private = lease_release_private_callback,
427 * Initialize a lease, use the default lock manager operations
429 static int lease_init(struct file *filp, int type, struct file_lock *fl)
431 fl->fl_owner = current->files;
432 fl->fl_pid = current->tgid;
435 fl->fl_flags = FL_LEASE;
436 if (assign_type(fl, type) != 0) {
441 fl->fl_end = OFFSET_MAX;
443 fl->fl_lmops = &lease_manager_ops;
447 /* Allocate a file_lock initialised to this type of lease */
448 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
450 struct file_lock *fl = locks_alloc_lock();
456 fl->fl_xid = vx_current_xid();
458 error = lease_init(filp, type, fl);
465 /* Check if two locks overlap each other.
467 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
469 return ((fl1->fl_end >= fl2->fl_start) &&
470 (fl2->fl_end >= fl1->fl_start));
474 * Check whether two locks have the same owner.
477 posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
479 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
480 return fl2->fl_lmops == fl1->fl_lmops &&
481 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
482 return fl1->fl_owner == fl2->fl_owner;
485 /* Remove waiter from blocker's block list.
486 * When blocker ends up pointing to itself then the list is empty.
488 static inline void __locks_delete_block(struct file_lock *waiter)
490 list_del_init(&waiter->fl_block);
491 list_del_init(&waiter->fl_link);
492 waiter->fl_next = NULL;
497 static void locks_delete_block(struct file_lock *waiter)
500 __locks_delete_block(waiter);
504 /* Insert waiter into blocker's block list.
505 * We use a circular list so that processes can be easily woken up in
506 * the order they blocked. The documentation doesn't require this but
507 * it seems like the reasonable thing to do.
509 static void locks_insert_block(struct file_lock *blocker,
510 struct file_lock *waiter)
512 if (!list_empty(&waiter->fl_block)) {
513 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
514 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
515 waiter->fl_start, waiter->fl_end, waiter->fl_type);
516 __locks_delete_block(waiter);
518 list_add_tail(&waiter->fl_block, &blocker->fl_block);
519 waiter->fl_next = blocker;
520 if (IS_POSIX(blocker))
521 list_add(&waiter->fl_link, &blocked_list);
524 /* Wake up processes blocked waiting for blocker.
525 * If told to wait then schedule the processes until the block list
526 * is empty, otherwise empty the block list ourselves.
528 static void locks_wake_up_blocks(struct file_lock *blocker)
530 while (!list_empty(&blocker->fl_block)) {
531 struct file_lock *waiter = list_entry(blocker->fl_block.next,
532 struct file_lock, fl_block);
533 __locks_delete_block(waiter);
534 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
535 waiter->fl_lmops->fl_notify(waiter);
537 wake_up(&waiter->fl_wait);
541 /* Insert file lock fl into an inode's lock list at the position indicated
542 * by pos. At the same time add the lock to the global file lock list.
544 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
546 list_add(&fl->fl_link, &file_lock_list);
548 /* insert into file's list */
552 if (fl->fl_ops && fl->fl_ops->fl_insert)
553 fl->fl_ops->fl_insert(fl);
557 * Delete a lock and then free it.
558 * Wake up processes that are blocked waiting for this lock,
559 * notify the FS that the lock has been cleared and
560 * finally free the lock.
562 static void locks_delete_lock(struct file_lock **thisfl_p)
564 struct file_lock *fl = *thisfl_p;
566 *thisfl_p = fl->fl_next;
568 list_del_init(&fl->fl_link);
570 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
571 if (fl->fl_fasync != NULL) {
572 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
573 fl->fl_fasync = NULL;
576 if (fl->fl_ops && fl->fl_ops->fl_remove)
577 fl->fl_ops->fl_remove(fl);
579 locks_wake_up_blocks(fl);
583 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
584 * checks for shared/exclusive status of overlapping locks.
586 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
588 if (sys_fl->fl_type == F_WRLCK)
590 if (caller_fl->fl_type == F_WRLCK)
595 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
596 * checking before calling the locks_conflict().
598 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
600 /* POSIX locks owned by the same process do not conflict with
603 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
606 /* Check whether they overlap */
607 if (!locks_overlap(caller_fl, sys_fl))
610 return (locks_conflict(caller_fl, sys_fl));
613 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
614 * checking before calling the locks_conflict().
616 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
618 /* FLOCK locks referring to the same filp do not conflict with
621 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
623 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
626 return (locks_conflict(caller_fl, sys_fl));
629 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
632 DECLARE_WAITQUEUE(wait, current);
634 __set_current_state(TASK_INTERRUPTIBLE);
635 add_wait_queue(fl_wait, &wait);
639 result = schedule_timeout(timeout);
640 if (signal_pending(current))
641 result = -ERESTARTSYS;
642 remove_wait_queue(fl_wait, &wait);
643 __set_current_state(TASK_RUNNING);
647 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
650 locks_insert_block(blocker, waiter);
651 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
652 __locks_delete_block(waiter);
657 posix_test_lock(struct file *filp, struct file_lock *fl)
659 struct file_lock *cfl;
662 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
665 if (posix_locks_conflict(cfl, fl))
673 EXPORT_SYMBOL(posix_test_lock);
675 /* This function tests for deadlock condition before putting a process to
676 * sleep. The detection scheme is no longer recursive. Recursive was neat,
677 * but dangerous - we risked stack corruption if the lock data was bad, or
678 * if the recursion was too deep for any other reason.
680 * We rely on the fact that a task can only be on one lock's wait queue
681 * at a time. When we find blocked_task on a wait queue we can re-search
682 * with blocked_task equal to that queue's owner, until either blocked_task
683 * isn't found, or blocked_task is found on a queue owned by my_task.
685 * Note: the above assumption may not be true when handling lock requests
686 * from a broken NFS client. But broken NFS clients have a lot more to
687 * worry about than proper deadlock detection anyway... --okir
689 int posix_locks_deadlock(struct file_lock *caller_fl,
690 struct file_lock *block_fl)
692 struct list_head *tmp;
695 if (posix_same_owner(caller_fl, block_fl))
697 list_for_each(tmp, &blocked_list) {
698 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
699 if (posix_same_owner(fl, block_fl)) {
708 EXPORT_SYMBOL(posix_locks_deadlock);
710 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
711 * at the head of the list, but that's secret knowledge known only to
712 * flock_lock_file and posix_lock_file.
714 static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
716 struct file_lock **before;
717 struct inode * inode = filp->f_dentry->d_inode;
722 for_each_lock(inode, before) {
723 struct file_lock *fl = *before;
728 if (filp != fl->fl_file)
730 if (new_fl->fl_type == fl->fl_type)
733 locks_delete_lock(before);
738 if (new_fl->fl_type == F_UNLCK)
742 * If a higher-priority process was blocked on the old file lock,
743 * give it the opportunity to lock the file.
749 for_each_lock(inode, before) {
750 struct file_lock *fl = *before;
755 if (!flock_locks_conflict(new_fl, fl))
758 if (new_fl->fl_flags & FL_SLEEP) {
759 locks_insert_block(fl, new_fl);
763 locks_insert_lock(&inode->i_flock, new_fl);
771 EXPORT_SYMBOL(posix_lock_file);
773 static int __posix_lock_file(struct inode *inode, struct file_lock *request)
775 struct file_lock *fl;
776 struct file_lock *new_fl, *new_fl2;
777 struct file_lock *left = NULL;
778 struct file_lock *right = NULL;
779 struct file_lock **before;
780 int error, added = 0;
783 * We may need two file_lock structures for this operation,
784 * so we get them in advance to avoid races.
786 new_fl = locks_alloc_lock();
787 new_fl->fl_xid = vx_current_xid();
788 vx_locks_inc(new_fl);
789 new_fl2 = locks_alloc_lock();
790 new_fl2->fl_xid = vx_current_xid();
791 vx_locks_inc(new_fl2);
794 if (request->fl_type != F_UNLCK) {
795 for_each_lock(inode, before) {
796 struct file_lock *fl = *before;
799 if (!posix_locks_conflict(request, fl))
802 if (!(request->fl_flags & FL_SLEEP))
805 if (posix_locks_deadlock(request, fl))
808 locks_insert_block(fl, request);
813 /* If we're just looking for a conflict, we're done. */
815 if (request->fl_flags & FL_ACCESS)
818 error = -ENOLCK; /* "no luck" */
819 if (!(new_fl && new_fl2))
823 * We've allocated the new locks in advance, so there are no
824 * errors possible (and no blocking operations) from here on.
826 * Find the first old lock with the same owner as the new lock.
829 before = &inode->i_flock;
831 /* First skip locks owned by other processes. */
832 while ((fl = *before) && (!IS_POSIX(fl) ||
833 !posix_same_owner(request, fl))) {
834 before = &fl->fl_next;
837 /* Process locks with this owner. */
838 while ((fl = *before) && posix_same_owner(request, fl)) {
839 /* Detect adjacent or overlapping regions (if same lock type)
841 if (request->fl_type == fl->fl_type) {
842 if (fl->fl_end < request->fl_start - 1)
844 /* If the next lock in the list has entirely bigger
845 * addresses than the new one, insert the lock here.
847 if (fl->fl_start > request->fl_end + 1)
850 /* If we come here, the new and old lock are of the
851 * same type and adjacent or overlapping. Make one
852 * lock yielding from the lower start address of both
853 * locks to the higher end address.
855 if (fl->fl_start > request->fl_start)
856 fl->fl_start = request->fl_start;
858 request->fl_start = fl->fl_start;
859 if (fl->fl_end < request->fl_end)
860 fl->fl_end = request->fl_end;
862 request->fl_end = fl->fl_end;
864 locks_delete_lock(before);
871 /* Processing for different lock types is a bit
874 if (fl->fl_end < request->fl_start)
876 if (fl->fl_start > request->fl_end)
878 if (request->fl_type == F_UNLCK)
880 if (fl->fl_start < request->fl_start)
882 /* If the next lock in the list has a higher end
883 * address than the new one, insert the new one here.
885 if (fl->fl_end > request->fl_end) {
889 if (fl->fl_start >= request->fl_start) {
890 /* The new lock completely replaces an old
891 * one (This may happen several times).
894 locks_delete_lock(before);
897 /* Replace the old lock with the new one.
898 * Wake up anybody waiting for the old one,
899 * as the change in lock type might satisfy
902 locks_wake_up_blocks(fl);
903 fl->fl_start = request->fl_start;
904 fl->fl_end = request->fl_end;
905 fl->fl_type = request->fl_type;
906 fl->fl_u = request->fl_u;
911 /* Go on to next lock.
914 before = &fl->fl_next;
919 if (request->fl_type == F_UNLCK)
921 locks_copy_lock(new_fl, request);
922 locks_insert_lock(before, new_fl);
927 /* The new lock breaks the old one in two pieces,
928 * so we have to use the second new lock.
932 locks_copy_lock(left, right);
933 locks_insert_lock(before, left);
935 right->fl_start = request->fl_end + 1;
936 locks_wake_up_blocks(right);
939 left->fl_end = request->fl_start - 1;
940 locks_wake_up_blocks(left);
945 * Free any unused locks.
948 locks_free_lock(new_fl);
950 locks_free_lock(new_fl2);
955 * posix_lock_file - Apply a POSIX-style lock to a file
956 * @filp: The file to apply the lock to
957 * @fl: The lock to be applied
959 * Add a POSIX style lock to a file.
960 * We merge adjacent & overlapping locks whenever possible.
961 * POSIX locks are sorted by owner task, then by starting address
963 int posix_lock_file(struct file *filp, struct file_lock *fl)
965 return __posix_lock_file(filp->f_dentry->d_inode, fl);
969 * posix_lock_file_wait - Apply a POSIX-style lock to a file
970 * @filp: The file to apply the lock to
971 * @fl: The lock to be applied
973 * Add a POSIX style lock to a file.
974 * We merge adjacent & overlapping locks whenever possible.
975 * POSIX locks are sorted by owner task, then by starting address
977 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
982 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
983 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
985 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
989 locks_delete_block(fl);
994 EXPORT_SYMBOL(posix_lock_file_wait);
997 * locks_mandatory_locked - Check for an active lock
998 * @inode: the file to check
1000 * Searches the inode's list of locks to find any POSIX locks which conflict.
1001 * This function is called from locks_verify_locked() only.
1003 int locks_mandatory_locked(struct inode *inode)
1005 fl_owner_t owner = current->files;
1006 struct file_lock *fl;
1009 * Search the lock list for this inode for any POSIX locks.
1012 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1015 if (fl->fl_owner != owner)
1019 return fl ? -EAGAIN : 0;
1023 * locks_mandatory_area - Check for a conflicting lock
1024 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1026 * @inode: the file to check
1027 * @filp: how the file was opened (if it was)
1028 * @offset: start of area to check
1029 * @count: length of area to check
1031 * Searches the inode's list of locks to find any POSIX locks which conflict.
1032 * This function is called from rw_verify_area() and
1033 * locks_verify_truncate().
1035 int locks_mandatory_area(int read_write, struct inode *inode,
1036 struct file *filp, loff_t offset,
1039 struct file_lock fl;
1042 locks_init_lock(&fl);
1043 fl.fl_owner = current->files;
1044 fl.fl_pid = current->tgid;
1046 fl.fl_flags = FL_POSIX | FL_ACCESS;
1047 if (filp && !(filp->f_flags & O_NONBLOCK))
1048 fl.fl_flags |= FL_SLEEP;
1049 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1050 fl.fl_start = offset;
1051 fl.fl_end = offset + count - 1;
1054 error = __posix_lock_file(inode, &fl);
1055 if (error != -EAGAIN)
1057 if (!(fl.fl_flags & FL_SLEEP))
1059 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1062 * If we've been sleeping someone might have
1063 * changed the permissions behind our back.
1065 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1069 locks_delete_block(&fl);
1076 EXPORT_SYMBOL(locks_mandatory_area);
1078 /* We already had a lease on this file; just change its type */
1079 static int lease_modify(struct file_lock **before, int arg)
1081 struct file_lock *fl = *before;
1082 int error = assign_type(fl, arg);
1086 locks_wake_up_blocks(fl);
1088 locks_delete_lock(before);
1092 static void time_out_leases(struct inode *inode)
1094 struct file_lock **before;
1095 struct file_lock *fl;
1097 before = &inode->i_flock;
1098 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1099 if ((fl->fl_break_time == 0)
1100 || time_before(jiffies, fl->fl_break_time)) {
1101 before = &fl->fl_next;
1104 printk(KERN_INFO "lease broken - owner pid = %d\n", fl->fl_pid);
1105 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1106 if (fl == *before) /* lease_modify may have freed fl */
1107 before = &fl->fl_next;
1112 * remove_lease - let time_out_leases remove the lease.
1113 * @@file_lock: the lease to remove
1115 void remove_lease(struct file_lock *fl)
1118 if (!fl || !IS_LEASE(fl))
1120 fl->fl_type = F_UNLCK | F_INPROGRESS;
1121 fl->fl_break_time = jiffies - 10;
1122 time_out_leases(fl->fl_file->f_dentry->d_inode);
1127 EXPORT_SYMBOL(remove_lease);
1130 * __break_lease - revoke all outstanding leases on file
1131 * @inode: the inode of the file to return
1132 * @mode: the open mode (read or write)
1134 * break_lease (inlined for speed) has checked there already
1135 * is a lease on this file. Leases are broken on a call to open()
1136 * or truncate(). This function can sleep unless you
1137 * specified %O_NONBLOCK to your open().
1139 int __break_lease(struct inode *inode, unsigned int mode)
1141 int error = 0, future;
1142 struct file_lock *new_fl, *flock;
1143 struct file_lock *fl;
1145 unsigned long break_time;
1146 int i_have_this_lease = 0;
1148 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1153 time_out_leases(inode);
1155 flock = inode->i_flock;
1156 if ((flock == NULL) || !IS_LEASE(flock))
1159 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1160 if (fl->fl_owner == current->files)
1161 i_have_this_lease = 1;
1163 if (mode & FMODE_WRITE) {
1164 /* If we want write access, we have to revoke any lease. */
1165 future = F_UNLCK | F_INPROGRESS;
1166 } else if (flock->fl_type & F_INPROGRESS) {
1167 /* If the lease is already being broken, we just leave it */
1168 future = flock->fl_type;
1169 } else if (flock->fl_type & F_WRLCK) {
1170 /* Downgrade the exclusive lease to a read-only lease. */
1171 future = F_RDLCK | F_INPROGRESS;
1173 /* the existing lease was read-only, so we can read too. */
1177 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1183 if (lease_break_time > 0) {
1184 break_time = jiffies + lease_break_time * HZ;
1185 if (break_time == 0)
1186 break_time++; /* so that 0 means no break time */
1189 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1190 if (fl->fl_type != future) {
1191 fl->fl_type = future;
1192 fl->fl_break_time = break_time;
1193 if (fl->fl_lmops && fl->fl_lmops->fl_break)
1194 fl->fl_lmops->fl_break(fl);
1195 else /* lease must have lmops break callback */
1200 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1201 error = -EWOULDBLOCK;
1206 break_time = flock->fl_break_time;
1207 if (break_time != 0) {
1208 break_time -= jiffies;
1209 if (break_time == 0)
1212 error = locks_block_on_timeout(flock, new_fl, break_time);
1215 time_out_leases(inode);
1216 /* Wait for the next lease that has not been broken yet */
1217 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1218 flock = flock->fl_next) {
1219 if (flock->fl_type & F_INPROGRESS)
1228 locks_free_lock(new_fl);
1232 EXPORT_SYMBOL(__break_lease);
1237 * @time: pointer to a timespec which will contain the last modified time
1239 * This is to force NFS clients to flush their caches for files with
1240 * exclusive leases. The justification is that if someone has an
1241 * exclusive lease, then they could be modifiying it.
1243 void lease_get_mtime(struct inode *inode, struct timespec *time)
1245 struct file_lock *flock = inode->i_flock;
1246 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1247 *time = current_fs_time(inode->i_sb);
1249 *time = inode->i_mtime;
1252 EXPORT_SYMBOL(lease_get_mtime);
1255 * fcntl_getlease - Enquire what lease is currently active
1258 * The value returned by this function will be one of
1259 * (if no lease break is pending):
1261 * %F_RDLCK to indicate a shared lease is held.
1263 * %F_WRLCK to indicate an exclusive lease is held.
1265 * %F_UNLCK to indicate no lease is held.
1267 * (if a lease break is pending):
1269 * %F_RDLCK to indicate an exclusive lease needs to be
1270 * changed to a shared lease (or removed).
1272 * %F_UNLCK to indicate the lease needs to be removed.
1274 * XXX: sfr & willy disagree over whether F_INPROGRESS
1275 * should be returned to userspace.
1277 int fcntl_getlease(struct file *filp)
1279 struct file_lock *fl;
1283 time_out_leases(filp->f_dentry->d_inode);
1284 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1286 if (fl->fl_file == filp) {
1287 type = fl->fl_type & ~F_INPROGRESS;
1296 * __setlease - sets a lease on an open file
1297 * @filp: file pointer
1298 * @arg: type of lease to obtain
1299 * @flp: input - file_lock to use, output - file_lock inserted
1301 * The (input) flp->fl_lmops->fl_break function is required
1304 * Called with kernel lock held.
1306 int __setlease(struct file *filp, long arg, struct file_lock **flp)
1308 struct file_lock *fl, **before, **my_before = NULL, *lease = *flp;
1309 struct dentry *dentry = filp->f_dentry;
1310 struct inode *inode = dentry->d_inode;
1311 int error, rdlease_count = 0, wrlease_count = 0;
1313 time_out_leases(inode);
1316 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1320 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1322 if ((arg == F_WRLCK)
1323 && ((atomic_read(&dentry->d_count) > 1)
1324 || (atomic_read(&inode->i_count) > 1)))
1328 * At this point, we know that if there is an exclusive
1329 * lease on this file, then we hold it on this filp
1330 * (otherwise our open of this file would have blocked).
1331 * And if we are trying to acquire an exclusive lease,
1332 * then the file is not open by anyone (including us)
1333 * except for this filp.
1335 for (before = &inode->i_flock;
1336 ((fl = *before) != NULL) && IS_LEASE(fl);
1337 before = &fl->fl_next) {
1338 if (fl->fl_file == filp)
1340 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1342 * Someone is in the process of opening this
1343 * file for writing so we may not take an
1344 * exclusive lease on it.
1351 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1352 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1355 if (my_before != NULL) {
1356 error = lease_modify(my_before, arg);
1368 error = lease_alloc(filp, arg, &fl);
1372 locks_copy_lock(fl, lease);
1374 locks_insert_lock(before, fl);
1382 * setlease - sets a lease on an open file
1383 * @filp: file pointer
1384 * @arg: type of lease to obtain
1385 * @lease: file_lock to use
1387 * Call this to establish a lease on the file.
1388 * The fl_lmops fl_break function is required by break_lease
1391 int setlease(struct file *filp, long arg, struct file_lock **lease)
1393 struct dentry *dentry = filp->f_dentry;
1394 struct inode *inode = dentry->d_inode;
1397 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1399 if (!S_ISREG(inode->i_mode))
1401 error = security_file_lock(filp, arg);
1406 error = __setlease(filp, arg, lease);
1412 EXPORT_SYMBOL(setlease);
1415 * fcntl_setlease - sets a lease on an open file
1416 * @fd: open file descriptor
1417 * @filp: file pointer
1418 * @arg: type of lease to obtain
1420 * Call this fcntl to establish a lease on the file.
1421 * Note that you also need to call %F_SETSIG to
1422 * receive a signal when the lease is broken.
1424 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1426 struct file_lock fl, *flp = &fl;
1427 struct dentry *dentry = filp->f_dentry;
1428 struct inode *inode = dentry->d_inode;
1431 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1433 if (!S_ISREG(inode->i_mode))
1435 error = security_file_lock(filp, arg);
1439 locks_init_lock(&fl);
1440 error = lease_init(filp, arg, &fl);
1446 error = __setlease(filp, arg, &flp);
1450 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1452 /* remove lease just inserted by __setlease */
1453 flp->fl_type = F_UNLCK | F_INPROGRESS;
1454 flp->fl_break_time = jiffies- 10;
1455 time_out_leases(inode);
1459 error = f_setown(filp, current->pid, 0);
1466 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1467 * @filp: The file to apply the lock to
1468 * @fl: The lock to be applied
1470 * Add a FLOCK style lock to a file.
1472 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1477 error = flock_lock_file(filp, fl);
1478 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1480 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1484 locks_delete_block(fl);
1490 EXPORT_SYMBOL(flock_lock_file_wait);
1493 * sys_flock: - flock() system call.
1494 * @fd: the file descriptor to lock.
1495 * @cmd: the type of lock to apply.
1497 * Apply a %FL_FLOCK style lock to an open file descriptor.
1498 * The @cmd can be one of
1500 * %LOCK_SH -- a shared lock.
1502 * %LOCK_EX -- an exclusive lock.
1504 * %LOCK_UN -- remove an existing lock.
1506 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1508 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1509 * processes read and write access respectively.
1511 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1514 struct file_lock *lock;
1515 int can_sleep, unlock;
1523 can_sleep = !(cmd & LOCK_NB);
1525 unlock = (cmd == LOCK_UN);
1527 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1530 error = flock_make_lock(filp, &lock, cmd);
1534 lock->fl_flags |= FL_SLEEP;
1536 error = security_file_lock(filp, cmd);
1540 if (filp->f_op && filp->f_op->flock)
1541 error = filp->f_op->flock(filp,
1542 (can_sleep) ? F_SETLKW : F_SETLK,
1545 error = flock_lock_file_wait(filp, lock);
1548 if (list_empty(&lock->fl_link)) {
1549 locks_free_lock(lock);
1558 /* Report the first existing lock that would conflict with l.
1559 * This implements the F_GETLK command of fcntl().
1561 int fcntl_getlk(struct file *filp, struct flock __user *l)
1563 struct file_lock *fl, file_lock;
1568 if (copy_from_user(&flock, l, sizeof(flock)))
1571 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1574 error = flock_to_posix_lock(filp, &file_lock, &flock);
1578 if (filp->f_op && filp->f_op->lock) {
1579 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1583 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1585 fl = posix_test_lock(filp, &file_lock);
1588 flock.l_type = F_UNLCK;
1590 flock.l_pid = fl->fl_pid;
1591 #if BITS_PER_LONG == 32
1593 * Make sure we can represent the posix lock via
1594 * legacy 32bit flock.
1597 if (fl->fl_start > OFFT_OFFSET_MAX)
1599 if ((fl->fl_end != OFFSET_MAX)
1600 && (fl->fl_end > OFFT_OFFSET_MAX))
1603 flock.l_start = fl->fl_start;
1604 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1605 fl->fl_end - fl->fl_start + 1;
1607 flock.l_type = fl->fl_type;
1610 if (!copy_to_user(l, &flock, sizeof(flock)))
1616 /* Apply the lock described by l to an open file descriptor.
1617 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1619 int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
1621 struct file_lock *file_lock = locks_alloc_lock();
1623 struct inode *inode;
1626 if (file_lock == NULL)
1629 file_lock->fl_xid = vx_current_xid();
1630 vx_locks_inc(file_lock);
1633 * This might block, so we do it before checking the inode.
1636 if (copy_from_user(&flock, l, sizeof(flock)))
1639 inode = filp->f_dentry->d_inode;
1641 /* Don't allow mandatory locks on files that may be memory mapped
1644 if (IS_MANDLOCK(inode) &&
1645 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1646 mapping_writably_mapped(filp->f_mapping)) {
1651 error = flock_to_posix_lock(filp, file_lock, &flock);
1654 if (cmd == F_SETLKW) {
1655 file_lock->fl_flags |= FL_SLEEP;
1659 switch (flock.l_type) {
1661 if (!(filp->f_mode & FMODE_READ))
1665 if (!(filp->f_mode & FMODE_WRITE))
1675 error = security_file_lock(filp, file_lock->fl_type);
1679 if (filp->f_op && filp->f_op->lock != NULL) {
1680 error = filp->f_op->lock(filp, cmd, file_lock);
1685 error = __posix_lock_file(inode, file_lock);
1686 if ((error != -EAGAIN) || (cmd == F_SETLK))
1688 error = wait_event_interruptible(file_lock->fl_wait,
1689 !file_lock->fl_next);
1693 locks_delete_block(file_lock);
1698 locks_free_lock(file_lock);
1702 #if BITS_PER_LONG == 32
1703 /* Report the first existing lock that would conflict with l.
1704 * This implements the F_GETLK command of fcntl().
1706 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1708 struct file_lock *fl, file_lock;
1709 struct flock64 flock;
1713 if (copy_from_user(&flock, l, sizeof(flock)))
1716 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1719 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1723 if (filp->f_op && filp->f_op->lock) {
1724 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1728 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1730 fl = posix_test_lock(filp, &file_lock);
1733 flock.l_type = F_UNLCK;
1735 flock.l_pid = fl->fl_pid;
1736 flock.l_start = fl->fl_start;
1737 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1738 fl->fl_end - fl->fl_start + 1;
1740 flock.l_type = fl->fl_type;
1743 if (!copy_to_user(l, &flock, sizeof(flock)))
1750 /* Apply the lock described by l to an open file descriptor.
1751 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1753 int fcntl_setlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
1755 struct file_lock *file_lock = locks_alloc_lock();
1756 struct flock64 flock;
1757 struct inode *inode;
1760 if (file_lock == NULL)
1763 file_lock->fl_xid = vx_current_xid();
1764 vx_locks_inc(file_lock);
1767 * This might block, so we do it before checking the inode.
1770 if (copy_from_user(&flock, l, sizeof(flock)))
1773 inode = filp->f_dentry->d_inode;
1775 /* Don't allow mandatory locks on files that may be memory mapped
1778 if (IS_MANDLOCK(inode) &&
1779 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1780 mapping_writably_mapped(filp->f_mapping)) {
1785 error = flock64_to_posix_lock(filp, file_lock, &flock);
1788 if (cmd == F_SETLKW64) {
1789 file_lock->fl_flags |= FL_SLEEP;
1793 switch (flock.l_type) {
1795 if (!(filp->f_mode & FMODE_READ))
1799 if (!(filp->f_mode & FMODE_WRITE))
1809 error = security_file_lock(filp, file_lock->fl_type);
1813 if (filp->f_op && filp->f_op->lock != NULL) {
1814 error = filp->f_op->lock(filp, cmd, file_lock);
1819 error = __posix_lock_file(inode, file_lock);
1820 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1822 error = wait_event_interruptible(file_lock->fl_wait,
1823 !file_lock->fl_next);
1827 locks_delete_block(file_lock);
1832 locks_free_lock(file_lock);
1835 #endif /* BITS_PER_LONG == 32 */
1838 * This function is called when the file is being removed
1839 * from the task's fd array. POSIX locks belonging to this task
1840 * are deleted at this time.
1842 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1844 struct file_lock lock, **before;
1847 * If there are no locks held on this file, we don't need to call
1848 * posix_lock_file(). Another process could be setting a lock on this
1849 * file at the same time, but we wouldn't remove that lock anyway.
1851 before = &filp->f_dentry->d_inode->i_flock;
1852 if (*before == NULL)
1855 lock.fl_type = F_UNLCK;
1856 lock.fl_flags = FL_POSIX;
1858 lock.fl_end = OFFSET_MAX;
1859 lock.fl_owner = owner;
1860 lock.fl_pid = current->tgid;
1861 lock.fl_file = filp;
1863 lock.fl_lmops = NULL;
1865 if (filp->f_op && filp->f_op->lock != NULL) {
1866 filp->f_op->lock(filp, F_SETLK, &lock);
1870 /* Can't use posix_lock_file here; we need to remove it no matter
1871 * which pid we have.
1874 while (*before != NULL) {
1875 struct file_lock *fl = *before;
1876 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1877 locks_delete_lock(before);
1880 before = &fl->fl_next;
1884 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1885 lock.fl_ops->fl_release_private(&lock);
1888 EXPORT_SYMBOL(locks_remove_posix);
1891 * This function is called on the last close of an open file.
1893 void locks_remove_flock(struct file *filp)
1895 struct inode * inode = filp->f_dentry->d_inode;
1896 struct file_lock *fl;
1897 struct file_lock **before;
1899 if (!inode->i_flock)
1902 if (filp->f_op && filp->f_op->flock) {
1903 struct file_lock fl = { .fl_flags = FL_FLOCK,
1904 .fl_type = F_UNLCK };
1905 filp->f_op->flock(filp, F_SETLKW, &fl);
1909 before = &inode->i_flock;
1911 while ((fl = *before) != NULL) {
1912 if (fl->fl_file == filp) {
1914 * We might have a POSIX lock that was created at the same time
1915 * the filp was closed for the last time. Just remove that too,
1916 * regardless of ownership, since nobody can own it.
1918 if (IS_FLOCK(fl) || IS_POSIX(fl)) {
1919 locks_delete_lock(before);
1923 lease_modify(before, F_UNLCK);
1929 before = &fl->fl_next;
1935 * posix_block_lock - blocks waiting for a file lock
1936 * @blocker: the lock which is blocking
1937 * @waiter: the lock which conflicts and has to wait
1939 * lockd needs to block waiting for locks.
1942 posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1944 locks_insert_block(blocker, waiter);
1947 EXPORT_SYMBOL(posix_block_lock);
1950 * posix_unblock_lock - stop waiting for a file lock
1951 * @filp: how the file was opened
1952 * @waiter: the lock which was waiting
1954 * lockd needs to block waiting for locks.
1957 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1960 * A remote machine may cancel the lock request after it's been
1961 * granted locally. If that happens, we need to delete the lock.
1964 if (waiter->fl_next) {
1965 __locks_delete_block(waiter);
1969 waiter->fl_type = F_UNLCK;
1970 posix_lock_file(filp, waiter);
1974 EXPORT_SYMBOL(posix_unblock_lock);
1976 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1978 struct inode *inode = NULL;
1980 if (fl->fl_file != NULL)
1981 inode = fl->fl_file->f_dentry->d_inode;
1983 out += sprintf(out, "%d:%s ", id, pfx);
1985 out += sprintf(out, "%6s %s ",
1986 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1987 (inode == NULL) ? "*NOINODE*" :
1988 (IS_MANDLOCK(inode) &&
1989 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1990 "MANDATORY" : "ADVISORY ");
1991 } else if (IS_FLOCK(fl)) {
1992 if (fl->fl_type & LOCK_MAND) {
1993 out += sprintf(out, "FLOCK MSNFS ");
1995 out += sprintf(out, "FLOCK ADVISORY ");
1997 } else if (IS_LEASE(fl)) {
1998 out += sprintf(out, "LEASE ");
1999 if (fl->fl_type & F_INPROGRESS)
2000 out += sprintf(out, "BREAKING ");
2001 else if (fl->fl_file)
2002 out += sprintf(out, "ACTIVE ");
2004 out += sprintf(out, "BREAKER ");
2006 out += sprintf(out, "UNKNOWN UNKNOWN ");
2008 if (fl->fl_type & LOCK_MAND) {
2009 out += sprintf(out, "%s ",
2010 (fl->fl_type & LOCK_READ)
2011 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2012 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2014 out += sprintf(out, "%s ",
2015 (fl->fl_type & F_INPROGRESS)
2016 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2017 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2020 #ifdef WE_CAN_BREAK_LSLK_NOW
2021 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2022 inode->i_sb->s_id, inode->i_ino);
2024 /* userspace relies on this representation of dev_t ;-( */
2025 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2026 MAJOR(inode->i_sb->s_dev),
2027 MINOR(inode->i_sb->s_dev), inode->i_ino);
2030 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2033 if (fl->fl_end == OFFSET_MAX)
2034 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2036 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2039 out += sprintf(out, "0 EOF\n");
2043 static void move_lock_status(char **p, off_t* pos, off_t offset)
2047 if(*pos >= offset) {
2048 /* the complete line is valid */
2053 if(*pos+len > offset) {
2054 /* use the second part of the line */
2055 int i = offset-*pos;
2056 memmove(*p,*p+i,len-i);
2061 /* discard the complete line */
2066 * get_locks_status - reports lock usage in /proc/locks
2067 * @buffer: address in userspace to write into
2069 * @offset: how far we are through the buffer
2070 * @length: how much to read
2073 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2075 struct list_head *tmp;
2081 list_for_each(tmp, &file_lock_list) {
2082 struct list_head *btmp;
2083 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2084 lock_get_status(q, fl, ++i, "");
2085 move_lock_status(&q, &pos, offset);
2087 if(pos >= offset+length)
2090 list_for_each(btmp, &fl->fl_block) {
2091 struct file_lock *bfl = list_entry(btmp,
2092 struct file_lock, fl_block);
2093 lock_get_status(q, bfl, i, " ->");
2094 move_lock_status(&q, &pos, offset);
2096 if(pos >= offset+length)
2103 if(q-buffer < length)
2109 * lock_may_read - checks that the region is free of locks
2110 * @inode: the inode that is being read
2111 * @start: the first byte to read
2112 * @len: the number of bytes to read
2114 * Emulates Windows locking requirements. Whole-file
2115 * mandatory locks (share modes) can prohibit a read and
2116 * byte-range POSIX locks can prohibit a read if they overlap.
2118 * N.B. this function is only ever called
2119 * from knfsd and ownership of locks is never checked.
2121 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2123 struct file_lock *fl;
2126 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2128 if (fl->fl_type == F_RDLCK)
2130 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2132 } else if (IS_FLOCK(fl)) {
2133 if (!(fl->fl_type & LOCK_MAND))
2135 if (fl->fl_type & LOCK_READ)
2146 EXPORT_SYMBOL(lock_may_read);
2149 * lock_may_write - checks that the region is free of locks
2150 * @inode: the inode that is being written
2151 * @start: the first byte to write
2152 * @len: the number of bytes to write
2154 * Emulates Windows locking requirements. Whole-file
2155 * mandatory locks (share modes) can prohibit a write and
2156 * byte-range POSIX locks can prohibit a write if they overlap.
2158 * N.B. this function is only ever called
2159 * from knfsd and ownership of locks is never checked.
2161 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2163 struct file_lock *fl;
2166 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2168 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2170 } else if (IS_FLOCK(fl)) {
2171 if (!(fl->fl_type & LOCK_MAND))
2173 if (fl->fl_type & LOCK_WRITE)
2184 EXPORT_SYMBOL(lock_may_write);
2186 static inline void __steal_locks(struct file *file, fl_owner_t from)
2188 struct inode *inode = file->f_dentry->d_inode;
2189 struct file_lock *fl = inode->i_flock;
2192 if (fl->fl_file == file && fl->fl_owner == from)
2193 fl->fl_owner = current->files;
2198 /* When getting ready for executing a binary, we make sure that current
2199 * has a files_struct on its own. Before dropping the old files_struct,
2200 * we take over ownership of all locks for all file descriptors we own.
2201 * Note that we may accidentally steal a lock for a file that a sibling
2202 * has created since the unshare_files() call.
2204 void steal_locks(fl_owner_t from)
2206 struct files_struct *files = current->files;
2217 if (i >= files->max_fdset || i >= files->max_fds)
2219 set = files->open_fds->fds_bits[j++];
2222 struct file *file = files->fd[i];
2224 __steal_locks(file, from);
2232 EXPORT_SYMBOL(steal_locks);
2234 static int __init filelock_init(void)
2236 filelock_cache = kmem_cache_create("file_lock_cache",
2237 sizeof(struct file_lock), 0, SLAB_PANIC,
2242 core_initcall(filelock_init);