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>
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 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)
178 if (waitqueue_active(&fl->fl_wait))
179 panic("Attempting to free lock with active wait queue");
181 if (!list_empty(&fl->fl_block))
182 panic("Attempting to free lock with active block list");
184 if (!list_empty(&fl->fl_link))
185 panic("Attempting to free lock on active lock list");
187 locks_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;
208 EXPORT_SYMBOL(locks_init_lock);
211 * Initialises the fields of the file lock which are invariant for
214 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
216 struct file_lock *lock = (struct file_lock *) foo;
218 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
219 SLAB_CTOR_CONSTRUCTOR)
222 locks_init_lock(lock);
225 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
228 if (fl->fl_ops->fl_copy_lock)
229 fl->fl_ops->fl_copy_lock(new, fl);
230 new->fl_ops = fl->fl_ops;
233 if (fl->fl_lmops->fl_copy_lock)
234 fl->fl_lmops->fl_copy_lock(new, fl);
235 new->fl_lmops = fl->fl_lmops;
240 * Initialize a new lock from an existing file_lock structure.
242 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
244 locks_release_private(new);
246 new->fl_owner = fl->fl_owner;
247 new->fl_pid = fl->fl_pid;
248 new->fl_file = fl->fl_file;
249 new->fl_flags = fl->fl_flags;
250 new->fl_type = fl->fl_type;
251 new->fl_start = fl->fl_start;
252 new->fl_end = fl->fl_end;
253 new->fl_ops = fl->fl_ops;
254 new->fl_lmops = fl->fl_lmops;
256 locks_copy_private(new, fl);
259 EXPORT_SYMBOL(locks_copy_lock);
261 static inline int flock_translate_cmd(int cmd) {
263 return cmd & (LOCK_MAND | LOCK_RW);
275 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
276 static int flock_make_lock(struct file *filp, struct file_lock **lock,
279 struct file_lock *fl;
280 int type = flock_translate_cmd(cmd);
284 fl = locks_alloc_lock();
289 fl->fl_pid = current->tgid;
290 fl->fl_flags = FL_FLOCK;
292 fl->fl_end = OFFSET_MAX;
298 static int assign_type(struct file_lock *fl, int type)
312 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
315 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
320 switch (l->l_whence) {
328 start = i_size_read(filp->f_dentry->d_inode);
334 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
335 POSIX-2001 defines it. */
339 fl->fl_end = OFFSET_MAX;
341 end = start + l->l_len - 1;
343 } else if (l->l_len < 0) {
350 fl->fl_start = start; /* we record the absolute position */
351 if (fl->fl_end < fl->fl_start)
354 fl->fl_owner = current->files;
355 fl->fl_pid = current->tgid;
357 fl->fl_flags = FL_POSIX;
361 return assign_type(fl, l->l_type);
364 #if BITS_PER_LONG == 32
365 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
370 switch (l->l_whence) {
378 start = i_size_read(filp->f_dentry->d_inode);
387 fl->fl_end = OFFSET_MAX;
389 fl->fl_end = start + l->l_len - 1;
390 } else if (l->l_len < 0) {
391 fl->fl_end = start - 1;
396 fl->fl_start = start; /* we record the absolute position */
397 if (fl->fl_end < fl->fl_start)
400 fl->fl_owner = current->files;
401 fl->fl_pid = current->tgid;
403 fl->fl_flags = FL_POSIX;
411 fl->fl_type = l->l_type;
421 /* default lease lock manager operations */
422 static void lease_break_callback(struct file_lock *fl)
424 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
427 static void lease_release_private_callback(struct file_lock *fl)
432 f_delown(fl->fl_file);
433 fl->fl_file->f_owner.signum = 0;
436 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
438 return fl->fl_file == try->fl_file;
441 static struct lock_manager_operations lease_manager_ops = {
442 .fl_break = lease_break_callback,
443 .fl_release_private = lease_release_private_callback,
444 .fl_mylease = lease_mylease_callback,
445 .fl_change = lease_modify,
449 * Initialize a lease, use the default lock manager operations
451 static int lease_init(struct file *filp, int type, struct file_lock *fl)
453 fl->fl_owner = current->files;
454 fl->fl_pid = current->tgid;
457 fl->fl_flags = FL_LEASE;
458 if (assign_type(fl, type) != 0) {
463 fl->fl_end = OFFSET_MAX;
465 fl->fl_lmops = &lease_manager_ops;
469 /* Allocate a file_lock initialised to this type of lease */
470 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
472 struct file_lock *fl = locks_alloc_lock();
478 error = lease_init(filp, type, fl);
485 /* Check if two locks overlap each other.
487 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
489 return ((fl1->fl_end >= fl2->fl_start) &&
490 (fl2->fl_end >= fl1->fl_start));
494 * Check whether two locks have the same owner.
496 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
498 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
499 return fl2->fl_lmops == fl1->fl_lmops &&
500 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
501 return fl1->fl_owner == fl2->fl_owner;
504 /* Remove waiter from blocker's block list.
505 * When blocker ends up pointing to itself then the list is empty.
507 static void __locks_delete_block(struct file_lock *waiter)
509 list_del_init(&waiter->fl_block);
510 list_del_init(&waiter->fl_link);
511 waiter->fl_next = NULL;
516 static void locks_delete_block(struct file_lock *waiter)
519 __locks_delete_block(waiter);
523 /* Insert waiter into blocker's block list.
524 * We use a circular list so that processes can be easily woken up in
525 * the order they blocked. The documentation doesn't require this but
526 * it seems like the reasonable thing to do.
528 static void locks_insert_block(struct file_lock *blocker,
529 struct file_lock *waiter)
531 if (!list_empty(&waiter->fl_block)) {
532 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
533 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
534 waiter->fl_start, waiter->fl_end, waiter->fl_type);
535 __locks_delete_block(waiter);
537 list_add_tail(&waiter->fl_block, &blocker->fl_block);
538 waiter->fl_next = blocker;
539 if (IS_POSIX(blocker))
540 list_add(&waiter->fl_link, &blocked_list);
543 /* Wake up processes blocked waiting for blocker.
544 * If told to wait then schedule the processes until the block list
545 * is empty, otherwise empty the block list ourselves.
547 static void locks_wake_up_blocks(struct file_lock *blocker)
549 while (!list_empty(&blocker->fl_block)) {
550 struct file_lock *waiter = list_entry(blocker->fl_block.next,
551 struct file_lock, fl_block);
552 __locks_delete_block(waiter);
553 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
554 waiter->fl_lmops->fl_notify(waiter);
556 wake_up(&waiter->fl_wait);
560 /* Insert file lock fl into an inode's lock list at the position indicated
561 * by pos. At the same time add the lock to the global file lock list.
563 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
565 list_add(&fl->fl_link, &file_lock_list);
567 /* insert into file's list */
571 if (fl->fl_ops && fl->fl_ops->fl_insert)
572 fl->fl_ops->fl_insert(fl);
576 * Delete a lock and then free it.
577 * Wake up processes that are blocked waiting for this lock,
578 * notify the FS that the lock has been cleared and
579 * finally free the lock.
581 static void locks_delete_lock(struct file_lock **thisfl_p)
583 struct file_lock *fl = *thisfl_p;
585 *thisfl_p = fl->fl_next;
587 list_del_init(&fl->fl_link);
589 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
590 if (fl->fl_fasync != NULL) {
591 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
592 fl->fl_fasync = NULL;
595 if (fl->fl_ops && fl->fl_ops->fl_remove)
596 fl->fl_ops->fl_remove(fl);
598 locks_wake_up_blocks(fl);
602 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
603 * checks for shared/exclusive status of overlapping locks.
605 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
607 if (sys_fl->fl_type == F_WRLCK)
609 if (caller_fl->fl_type == F_WRLCK)
614 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
615 * checking before calling the locks_conflict().
617 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
619 /* POSIX locks owned by the same process do not conflict with
622 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
625 /* Check whether they overlap */
626 if (!locks_overlap(caller_fl, sys_fl))
629 return (locks_conflict(caller_fl, sys_fl));
632 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
633 * checking before calling the locks_conflict().
635 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
637 /* FLOCK locks referring to the same filp do not conflict with
640 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
642 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
645 return (locks_conflict(caller_fl, sys_fl));
648 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
651 DECLARE_WAITQUEUE(wait, current);
653 __set_current_state(TASK_INTERRUPTIBLE);
654 add_wait_queue(fl_wait, &wait);
658 result = schedule_timeout(timeout);
659 if (signal_pending(current))
660 result = -ERESTARTSYS;
661 remove_wait_queue(fl_wait, &wait);
662 __set_current_state(TASK_RUNNING);
666 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
669 locks_insert_block(blocker, waiter);
670 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
671 __locks_delete_block(waiter);
676 posix_test_lock(struct file *filp, struct file_lock *fl)
678 struct file_lock *cfl;
681 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
684 if (posix_locks_conflict(cfl, fl))
692 EXPORT_SYMBOL(posix_test_lock);
694 /* This function tests for deadlock condition before putting a process to
695 * sleep. The detection scheme is no longer recursive. Recursive was neat,
696 * but dangerous - we risked stack corruption if the lock data was bad, or
697 * if the recursion was too deep for any other reason.
699 * We rely on the fact that a task can only be on one lock's wait queue
700 * at a time. When we find blocked_task on a wait queue we can re-search
701 * with blocked_task equal to that queue's owner, until either blocked_task
702 * isn't found, or blocked_task is found on a queue owned by my_task.
704 * Note: the above assumption may not be true when handling lock requests
705 * from a broken NFS client. But broken NFS clients have a lot more to
706 * worry about than proper deadlock detection anyway... --okir
708 int posix_locks_deadlock(struct file_lock *caller_fl,
709 struct file_lock *block_fl)
711 struct list_head *tmp;
714 if (posix_same_owner(caller_fl, block_fl))
716 list_for_each(tmp, &blocked_list) {
717 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
718 if (posix_same_owner(fl, block_fl)) {
727 EXPORT_SYMBOL(posix_locks_deadlock);
729 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
730 * at the head of the list, but that's secret knowledge known only to
731 * flock_lock_file and posix_lock_file.
733 static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
735 struct file_lock **before;
736 struct inode * inode = filp->f_dentry->d_inode;
741 for_each_lock(inode, before) {
742 struct file_lock *fl = *before;
747 if (filp != fl->fl_file)
749 if (new_fl->fl_type == fl->fl_type)
752 locks_delete_lock(before);
757 if (new_fl->fl_type == F_UNLCK)
761 * If a higher-priority process was blocked on the old file lock,
762 * give it the opportunity to lock the file.
768 for_each_lock(inode, before) {
769 struct file_lock *fl = *before;
774 if (!flock_locks_conflict(new_fl, fl))
777 if (new_fl->fl_flags & FL_SLEEP) {
778 locks_insert_block(fl, new_fl);
782 locks_insert_lock(&inode->i_flock, new_fl);
790 EXPORT_SYMBOL(posix_lock_file);
792 static int __posix_lock_file(struct inode *inode, struct file_lock *request)
794 struct file_lock *fl;
795 struct file_lock *new_fl, *new_fl2;
796 struct file_lock *left = NULL;
797 struct file_lock *right = NULL;
798 struct file_lock **before;
799 int error, added = 0;
802 * We may need two file_lock structures for this operation,
803 * so we get them in advance to avoid races.
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
809 if (request->fl_type != F_UNLCK) {
810 for_each_lock(inode, before) {
811 struct file_lock *fl = *before;
814 if (!posix_locks_conflict(request, fl))
817 if (!(request->fl_flags & FL_SLEEP))
820 if (posix_locks_deadlock(request, fl))
823 locks_insert_block(fl, request);
828 /* If we're just looking for a conflict, we're done. */
830 if (request->fl_flags & FL_ACCESS)
833 error = -ENOLCK; /* "no luck" */
834 if (!(new_fl && new_fl2))
838 * We've allocated the new locks in advance, so there are no
839 * errors possible (and no blocking operations) from here on.
841 * Find the first old lock with the same owner as the new lock.
844 before = &inode->i_flock;
846 /* First skip locks owned by other processes. */
847 while ((fl = *before) && (!IS_POSIX(fl) ||
848 !posix_same_owner(request, fl))) {
849 before = &fl->fl_next;
852 /* Process locks with this owner. */
853 while ((fl = *before) && posix_same_owner(request, fl)) {
854 /* Detect adjacent or overlapping regions (if same lock type)
856 if (request->fl_type == fl->fl_type) {
857 /* In all comparisons of start vs end, use
858 * "start - 1" rather than "end + 1". If end
859 * is OFFSET_MAX, end + 1 will become negative.
861 if (fl->fl_end < request->fl_start - 1)
863 /* If the next lock in the list has entirely bigger
864 * addresses than the new one, insert the lock here.
866 if (fl->fl_start - 1 > request->fl_end)
869 /* If we come here, the new and old lock are of the
870 * same type and adjacent or overlapping. Make one
871 * lock yielding from the lower start address of both
872 * locks to the higher end address.
874 if (fl->fl_start > request->fl_start)
875 fl->fl_start = request->fl_start;
877 request->fl_start = fl->fl_start;
878 if (fl->fl_end < request->fl_end)
879 fl->fl_end = request->fl_end;
881 request->fl_end = fl->fl_end;
883 locks_delete_lock(before);
890 /* Processing for different lock types is a bit
893 if (fl->fl_end < request->fl_start)
895 if (fl->fl_start > request->fl_end)
897 if (request->fl_type == F_UNLCK)
899 if (fl->fl_start < request->fl_start)
901 /* If the next lock in the list has a higher end
902 * address than the new one, insert the new one here.
904 if (fl->fl_end > request->fl_end) {
908 if (fl->fl_start >= request->fl_start) {
909 /* The new lock completely replaces an old
910 * one (This may happen several times).
913 locks_delete_lock(before);
916 /* Replace the old lock with the new one.
917 * Wake up anybody waiting for the old one,
918 * as the change in lock type might satisfy
921 locks_wake_up_blocks(fl);
922 fl->fl_start = request->fl_start;
923 fl->fl_end = request->fl_end;
924 fl->fl_type = request->fl_type;
925 locks_release_private(fl);
926 locks_copy_private(fl, request);
931 /* Go on to next lock.
934 before = &fl->fl_next;
939 if (request->fl_type == F_UNLCK)
941 locks_copy_lock(new_fl, request);
942 locks_insert_lock(before, new_fl);
947 /* The new lock breaks the old one in two pieces,
948 * so we have to use the second new lock.
952 locks_copy_lock(left, right);
953 locks_insert_lock(before, left);
955 right->fl_start = request->fl_end + 1;
956 locks_wake_up_blocks(right);
959 left->fl_end = request->fl_start - 1;
960 locks_wake_up_blocks(left);
965 * Free any unused locks.
968 locks_free_lock(new_fl);
970 locks_free_lock(new_fl2);
975 * posix_lock_file - Apply a POSIX-style lock to a file
976 * @filp: The file to apply the lock to
977 * @fl: The lock to be applied
979 * Add a POSIX style lock to a file.
980 * We merge adjacent & overlapping locks whenever possible.
981 * POSIX locks are sorted by owner task, then by starting address
983 int posix_lock_file(struct file *filp, struct file_lock *fl)
985 return __posix_lock_file(filp->f_dentry->d_inode, fl);
989 * posix_lock_file_wait - Apply a POSIX-style lock to a file
990 * @filp: The file to apply the lock to
991 * @fl: The lock to be applied
993 * Add a POSIX style lock to a file.
994 * We merge adjacent & overlapping locks whenever possible.
995 * POSIX locks are sorted by owner task, then by starting address
997 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1002 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
1003 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1005 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1009 locks_delete_block(fl);
1014 EXPORT_SYMBOL(posix_lock_file_wait);
1017 * locks_mandatory_locked - Check for an active lock
1018 * @inode: the file to check
1020 * Searches the inode's list of locks to find any POSIX locks which conflict.
1021 * This function is called from locks_verify_locked() only.
1023 int locks_mandatory_locked(struct inode *inode)
1025 fl_owner_t owner = current->files;
1026 struct file_lock *fl;
1029 * Search the lock list for this inode for any POSIX locks.
1032 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1035 if (fl->fl_owner != owner)
1039 return fl ? -EAGAIN : 0;
1043 * locks_mandatory_area - Check for a conflicting lock
1044 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1046 * @inode: the file to check
1047 * @filp: how the file was opened (if it was)
1048 * @offset: start of area to check
1049 * @count: length of area to check
1051 * Searches the inode's list of locks to find any POSIX locks which conflict.
1052 * This function is called from rw_verify_area() and
1053 * locks_verify_truncate().
1055 int locks_mandatory_area(int read_write, struct inode *inode,
1056 struct file *filp, loff_t offset,
1059 struct file_lock fl;
1062 locks_init_lock(&fl);
1063 fl.fl_owner = current->files;
1064 fl.fl_pid = current->tgid;
1066 fl.fl_flags = FL_POSIX | FL_ACCESS;
1067 if (filp && !(filp->f_flags & O_NONBLOCK))
1068 fl.fl_flags |= FL_SLEEP;
1069 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1070 fl.fl_start = offset;
1071 fl.fl_end = offset + count - 1;
1074 error = __posix_lock_file(inode, &fl);
1075 if (error != -EAGAIN)
1077 if (!(fl.fl_flags & FL_SLEEP))
1079 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1082 * If we've been sleeping someone might have
1083 * changed the permissions behind our back.
1085 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1089 locks_delete_block(&fl);
1096 EXPORT_SYMBOL(locks_mandatory_area);
1098 /* We already had a lease on this file; just change its type */
1099 int lease_modify(struct file_lock **before, int arg)
1101 struct file_lock *fl = *before;
1102 int error = assign_type(fl, arg);
1106 locks_wake_up_blocks(fl);
1108 locks_delete_lock(before);
1112 EXPORT_SYMBOL(lease_modify);
1114 static void time_out_leases(struct inode *inode)
1116 struct file_lock **before;
1117 struct file_lock *fl;
1119 before = &inode->i_flock;
1120 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1121 if ((fl->fl_break_time == 0)
1122 || time_before(jiffies, fl->fl_break_time)) {
1123 before = &fl->fl_next;
1126 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1127 if (fl == *before) /* lease_modify may have freed fl */
1128 before = &fl->fl_next;
1133 * __break_lease - revoke all outstanding leases on file
1134 * @inode: the inode of the file to return
1135 * @mode: the open mode (read or write)
1137 * break_lease (inlined for speed) has checked there already
1138 * is a lease on this file. Leases are broken on a call to open()
1139 * or truncate(). This function can sleep unless you
1140 * specified %O_NONBLOCK to your open().
1142 int __break_lease(struct inode *inode, unsigned int mode)
1144 int error = 0, future;
1145 struct file_lock *new_fl, *flock;
1146 struct file_lock *fl;
1148 unsigned long break_time;
1149 int i_have_this_lease = 0;
1151 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1156 time_out_leases(inode);
1158 flock = inode->i_flock;
1159 if ((flock == NULL) || !IS_LEASE(flock))
1162 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1163 if (fl->fl_owner == current->files)
1164 i_have_this_lease = 1;
1166 if (mode & FMODE_WRITE) {
1167 /* If we want write access, we have to revoke any lease. */
1168 future = F_UNLCK | F_INPROGRESS;
1169 } else if (flock->fl_type & F_INPROGRESS) {
1170 /* If the lease is already being broken, we just leave it */
1171 future = flock->fl_type;
1172 } else if (flock->fl_type & F_WRLCK) {
1173 /* Downgrade the exclusive lease to a read-only lease. */
1174 future = F_RDLCK | F_INPROGRESS;
1176 /* the existing lease was read-only, so we can read too. */
1180 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1186 if (lease_break_time > 0) {
1187 break_time = jiffies + lease_break_time * HZ;
1188 if (break_time == 0)
1189 break_time++; /* so that 0 means no break time */
1192 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1193 if (fl->fl_type != future) {
1194 fl->fl_type = future;
1195 fl->fl_break_time = break_time;
1196 /* lease must have lmops break callback */
1197 fl->fl_lmops->fl_break(fl);
1201 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1202 error = -EWOULDBLOCK;
1207 break_time = flock->fl_break_time;
1208 if (break_time != 0) {
1209 break_time -= jiffies;
1210 if (break_time == 0)
1213 error = locks_block_on_timeout(flock, new_fl, break_time);
1216 time_out_leases(inode);
1217 /* Wait for the next lease that has not been broken yet */
1218 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1219 flock = flock->fl_next) {
1220 if (flock->fl_type & F_INPROGRESS)
1229 locks_free_lock(new_fl);
1233 EXPORT_SYMBOL(__break_lease);
1238 * @time: pointer to a timespec which will contain the last modified time
1240 * This is to force NFS clients to flush their caches for files with
1241 * exclusive leases. The justification is that if someone has an
1242 * exclusive lease, then they could be modifiying it.
1244 void lease_get_mtime(struct inode *inode, struct timespec *time)
1246 struct file_lock *flock = inode->i_flock;
1247 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1248 *time = current_fs_time(inode->i_sb);
1250 *time = inode->i_mtime;
1253 EXPORT_SYMBOL(lease_get_mtime);
1256 * fcntl_getlease - Enquire what lease is currently active
1259 * The value returned by this function will be one of
1260 * (if no lease break is pending):
1262 * %F_RDLCK to indicate a shared lease is held.
1264 * %F_WRLCK to indicate an exclusive lease is held.
1266 * %F_UNLCK to indicate no lease is held.
1268 * (if a lease break is pending):
1270 * %F_RDLCK to indicate an exclusive lease needs to be
1271 * changed to a shared lease (or removed).
1273 * %F_UNLCK to indicate the lease needs to be removed.
1275 * XXX: sfr & willy disagree over whether F_INPROGRESS
1276 * should be returned to userspace.
1278 int fcntl_getlease(struct file *filp)
1280 struct file_lock *fl;
1284 time_out_leases(filp->f_dentry->d_inode);
1285 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1287 if (fl->fl_file == filp) {
1288 type = fl->fl_type & ~F_INPROGRESS;
1297 * __setlease - sets a lease on an open file
1298 * @filp: file pointer
1299 * @arg: type of lease to obtain
1300 * @flp: input - file_lock to use, output - file_lock inserted
1302 * The (input) flp->fl_lmops->fl_break function is required
1305 * Called with kernel lock held.
1307 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1309 struct file_lock *fl, **before, **my_before = NULL, *lease;
1310 struct dentry *dentry = filp->f_dentry;
1311 struct inode *inode = dentry->d_inode;
1312 int error, rdlease_count = 0, wrlease_count = 0;
1314 time_out_leases(inode);
1317 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1323 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1325 if ((arg == F_WRLCK)
1326 && ((atomic_read(&dentry->d_count) > 1)
1327 || (atomic_read(&inode->i_count) > 1)))
1331 * At this point, we know that if there is an exclusive
1332 * lease on this file, then we hold it on this filp
1333 * (otherwise our open of this file would have blocked).
1334 * And if we are trying to acquire an exclusive lease,
1335 * then the file is not open by anyone (including us)
1336 * except for this filp.
1338 for (before = &inode->i_flock;
1339 ((fl = *before) != NULL) && IS_LEASE(fl);
1340 before = &fl->fl_next) {
1341 if (lease->fl_lmops->fl_mylease(fl, lease))
1343 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1345 * Someone is in the process of opening this
1346 * file for writing so we may not take an
1347 * exclusive lease on it.
1354 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1355 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1358 if (my_before != NULL) {
1359 error = lease->fl_lmops->fl_change(my_before, arg);
1371 error = lease_alloc(filp, arg, &fl);
1375 locks_copy_lock(fl, lease);
1377 locks_insert_lock(before, fl);
1385 * setlease - sets a lease on an open file
1386 * @filp: file pointer
1387 * @arg: type of lease to obtain
1388 * @lease: file_lock to use
1390 * Call this to establish a lease on the file.
1391 * The fl_lmops fl_break function is required by break_lease
1394 int setlease(struct file *filp, long arg, struct file_lock **lease)
1396 struct dentry *dentry = filp->f_dentry;
1397 struct inode *inode = dentry->d_inode;
1400 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1402 if (!S_ISREG(inode->i_mode))
1404 error = security_file_lock(filp, arg);
1409 error = __setlease(filp, arg, lease);
1415 EXPORT_SYMBOL(setlease);
1418 * fcntl_setlease - sets a lease on an open file
1419 * @fd: open file descriptor
1420 * @filp: file pointer
1421 * @arg: type of lease to obtain
1423 * Call this fcntl to establish a lease on the file.
1424 * Note that you also need to call %F_SETSIG to
1425 * receive a signal when the lease is broken.
1427 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1429 struct file_lock fl, *flp = &fl;
1430 struct dentry *dentry = filp->f_dentry;
1431 struct inode *inode = dentry->d_inode;
1434 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1436 if (!S_ISREG(inode->i_mode))
1438 error = security_file_lock(filp, arg);
1442 locks_init_lock(&fl);
1443 error = lease_init(filp, arg, &fl);
1449 error = __setlease(filp, arg, &flp);
1450 if (error || arg == F_UNLCK)
1453 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1455 /* remove lease just inserted by __setlease */
1456 flp->fl_type = F_UNLCK | F_INPROGRESS;
1457 flp->fl_break_time = jiffies- 10;
1458 time_out_leases(inode);
1462 error = f_setown(filp, current->pid, 0);
1469 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1470 * @filp: The file to apply the lock to
1471 * @fl: The lock to be applied
1473 * Add a FLOCK style lock to a file.
1475 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1480 error = flock_lock_file(filp, fl);
1481 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1483 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1487 locks_delete_block(fl);
1493 EXPORT_SYMBOL(flock_lock_file_wait);
1496 * sys_flock: - flock() system call.
1497 * @fd: the file descriptor to lock.
1498 * @cmd: the type of lock to apply.
1500 * Apply a %FL_FLOCK style lock to an open file descriptor.
1501 * The @cmd can be one of
1503 * %LOCK_SH -- a shared lock.
1505 * %LOCK_EX -- an exclusive lock.
1507 * %LOCK_UN -- remove an existing lock.
1509 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1511 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1512 * processes read and write access respectively.
1514 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1517 struct file_lock *lock;
1518 int can_sleep, unlock;
1526 can_sleep = !(cmd & LOCK_NB);
1528 unlock = (cmd == LOCK_UN);
1530 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1533 error = flock_make_lock(filp, &lock, cmd);
1537 lock->fl_flags |= FL_SLEEP;
1539 error = security_file_lock(filp, cmd);
1543 if (filp->f_op && filp->f_op->flock)
1544 error = filp->f_op->flock(filp,
1545 (can_sleep) ? F_SETLKW : F_SETLK,
1548 error = flock_lock_file_wait(filp, lock);
1551 if (list_empty(&lock->fl_link)) {
1552 locks_free_lock(lock);
1561 /* Report the first existing lock that would conflict with l.
1562 * This implements the F_GETLK command of fcntl().
1564 int fcntl_getlk(struct file *filp, struct flock __user *l)
1566 struct file_lock *fl, file_lock;
1571 if (copy_from_user(&flock, l, sizeof(flock)))
1574 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1577 error = flock_to_posix_lock(filp, &file_lock, &flock);
1581 if (filp->f_op && filp->f_op->lock) {
1582 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1583 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1584 file_lock.fl_ops->fl_release_private(&file_lock);
1588 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1590 fl = posix_test_lock(filp, &file_lock);
1593 flock.l_type = F_UNLCK;
1595 flock.l_pid = fl->fl_pid;
1596 #if BITS_PER_LONG == 32
1598 * Make sure we can represent the posix lock via
1599 * legacy 32bit flock.
1602 if (fl->fl_start > OFFT_OFFSET_MAX)
1604 if ((fl->fl_end != OFFSET_MAX)
1605 && (fl->fl_end > OFFT_OFFSET_MAX))
1608 flock.l_start = fl->fl_start;
1609 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1610 fl->fl_end - fl->fl_start + 1;
1612 flock.l_type = fl->fl_type;
1615 if (!copy_to_user(l, &flock, sizeof(flock)))
1621 /* Apply the lock described by l to an open file descriptor.
1622 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1624 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1625 struct flock __user *l)
1627 struct file_lock *file_lock = locks_alloc_lock();
1629 struct inode *inode;
1632 if (file_lock == NULL)
1636 * This might block, so we do it before checking the inode.
1639 if (copy_from_user(&flock, l, sizeof(flock)))
1642 inode = filp->f_dentry->d_inode;
1644 /* Don't allow mandatory locks on files that may be memory mapped
1647 if (IS_MANDLOCK(inode) &&
1648 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1649 mapping_writably_mapped(filp->f_mapping)) {
1655 error = flock_to_posix_lock(filp, file_lock, &flock);
1658 if (cmd == F_SETLKW) {
1659 file_lock->fl_flags |= FL_SLEEP;
1663 switch (flock.l_type) {
1665 if (!(filp->f_mode & FMODE_READ))
1669 if (!(filp->f_mode & FMODE_WRITE))
1679 error = security_file_lock(filp, file_lock->fl_type);
1683 if (filp->f_op && filp->f_op->lock != NULL)
1684 error = filp->f_op->lock(filp, cmd, file_lock);
1687 error = __posix_lock_file(inode, file_lock);
1688 if ((error != -EAGAIN) || (cmd == F_SETLK))
1690 error = wait_event_interruptible(file_lock->fl_wait,
1691 !file_lock->fl_next);
1695 locks_delete_block(file_lock);
1701 * Attempt to detect a close/fcntl race and recover by
1702 * releasing the lock that was just acquired.
1704 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1705 flock.l_type = F_UNLCK;
1710 locks_free_lock(file_lock);
1714 #if BITS_PER_LONG == 32
1715 /* Report the first existing lock that would conflict with l.
1716 * This implements the F_GETLK command of fcntl().
1718 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1720 struct file_lock *fl, file_lock;
1721 struct flock64 flock;
1725 if (copy_from_user(&flock, l, sizeof(flock)))
1728 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1731 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1735 if (filp->f_op && filp->f_op->lock) {
1736 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1737 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1738 file_lock.fl_ops->fl_release_private(&file_lock);
1742 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1744 fl = posix_test_lock(filp, &file_lock);
1747 flock.l_type = F_UNLCK;
1749 flock.l_pid = fl->fl_pid;
1750 flock.l_start = fl->fl_start;
1751 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1752 fl->fl_end - fl->fl_start + 1;
1754 flock.l_type = fl->fl_type;
1757 if (!copy_to_user(l, &flock, sizeof(flock)))
1764 /* Apply the lock described by l to an open file descriptor.
1765 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1767 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1768 struct flock64 __user *l)
1770 struct file_lock *file_lock = locks_alloc_lock();
1771 struct flock64 flock;
1772 struct inode *inode;
1775 if (file_lock == NULL)
1779 * This might block, so we do it before checking the inode.
1782 if (copy_from_user(&flock, l, sizeof(flock)))
1785 inode = filp->f_dentry->d_inode;
1787 /* Don't allow mandatory locks on files that may be memory mapped
1790 if (IS_MANDLOCK(inode) &&
1791 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1792 mapping_writably_mapped(filp->f_mapping)) {
1798 error = flock64_to_posix_lock(filp, file_lock, &flock);
1801 if (cmd == F_SETLKW64) {
1802 file_lock->fl_flags |= FL_SLEEP;
1806 switch (flock.l_type) {
1808 if (!(filp->f_mode & FMODE_READ))
1812 if (!(filp->f_mode & FMODE_WRITE))
1822 error = security_file_lock(filp, file_lock->fl_type);
1826 if (filp->f_op && filp->f_op->lock != NULL)
1827 error = filp->f_op->lock(filp, cmd, file_lock);
1830 error = __posix_lock_file(inode, file_lock);
1831 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1833 error = wait_event_interruptible(file_lock->fl_wait,
1834 !file_lock->fl_next);
1838 locks_delete_block(file_lock);
1844 * Attempt to detect a close/fcntl race and recover by
1845 * releasing the lock that was just acquired.
1847 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1848 flock.l_type = F_UNLCK;
1853 locks_free_lock(file_lock);
1856 #endif /* BITS_PER_LONG == 32 */
1859 * This function is called when the file is being removed
1860 * from the task's fd array. POSIX locks belonging to this task
1861 * are deleted at this time.
1863 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1865 struct file_lock lock, **before;
1868 * If there are no locks held on this file, we don't need to call
1869 * posix_lock_file(). Another process could be setting a lock on this
1870 * file at the same time, but we wouldn't remove that lock anyway.
1872 before = &filp->f_dentry->d_inode->i_flock;
1873 if (*before == NULL)
1876 lock.fl_type = F_UNLCK;
1877 lock.fl_flags = FL_POSIX;
1879 lock.fl_end = OFFSET_MAX;
1880 lock.fl_owner = owner;
1881 lock.fl_pid = current->tgid;
1882 lock.fl_file = filp;
1884 lock.fl_lmops = NULL;
1886 if (filp->f_op && filp->f_op->lock != NULL) {
1887 filp->f_op->lock(filp, F_SETLK, &lock);
1891 /* Can't use posix_lock_file here; we need to remove it no matter
1892 * which pid we have.
1895 while (*before != NULL) {
1896 struct file_lock *fl = *before;
1897 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1898 locks_delete_lock(before);
1901 before = &fl->fl_next;
1905 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1906 lock.fl_ops->fl_release_private(&lock);
1909 EXPORT_SYMBOL(locks_remove_posix);
1912 * This function is called on the last close of an open file.
1914 void locks_remove_flock(struct file *filp)
1916 struct inode * inode = filp->f_dentry->d_inode;
1917 struct file_lock *fl;
1918 struct file_lock **before;
1920 if (!inode->i_flock)
1923 if (filp->f_op && filp->f_op->flock) {
1924 struct file_lock fl = {
1925 .fl_pid = current->tgid,
1927 .fl_flags = FL_FLOCK,
1929 .fl_end = OFFSET_MAX,
1931 filp->f_op->flock(filp, F_SETLKW, &fl);
1932 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1933 fl.fl_ops->fl_release_private(&fl);
1937 before = &inode->i_flock;
1939 while ((fl = *before) != NULL) {
1940 if (fl->fl_file == filp) {
1942 locks_delete_lock(before);
1946 lease_modify(before, F_UNLCK);
1952 before = &fl->fl_next;
1958 * posix_block_lock - blocks waiting for a file lock
1959 * @blocker: the lock which is blocking
1960 * @waiter: the lock which conflicts and has to wait
1962 * lockd needs to block waiting for locks.
1965 posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1967 locks_insert_block(blocker, waiter);
1970 EXPORT_SYMBOL(posix_block_lock);
1973 * posix_unblock_lock - stop waiting for a file lock
1974 * @filp: how the file was opened
1975 * @waiter: the lock which was waiting
1977 * lockd needs to block waiting for locks.
1980 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1985 if (waiter->fl_next)
1986 __locks_delete_block(waiter);
1993 EXPORT_SYMBOL(posix_unblock_lock);
1995 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1997 struct inode *inode = NULL;
1999 if (fl->fl_file != NULL)
2000 inode = fl->fl_file->f_dentry->d_inode;
2002 out += sprintf(out, "%d:%s ", id, pfx);
2004 out += sprintf(out, "%6s %s ",
2005 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2006 (inode == NULL) ? "*NOINODE*" :
2007 (IS_MANDLOCK(inode) &&
2008 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2009 "MANDATORY" : "ADVISORY ");
2010 } else if (IS_FLOCK(fl)) {
2011 if (fl->fl_type & LOCK_MAND) {
2012 out += sprintf(out, "FLOCK MSNFS ");
2014 out += sprintf(out, "FLOCK ADVISORY ");
2016 } else if (IS_LEASE(fl)) {
2017 out += sprintf(out, "LEASE ");
2018 if (fl->fl_type & F_INPROGRESS)
2019 out += sprintf(out, "BREAKING ");
2020 else if (fl->fl_file)
2021 out += sprintf(out, "ACTIVE ");
2023 out += sprintf(out, "BREAKER ");
2025 out += sprintf(out, "UNKNOWN UNKNOWN ");
2027 if (fl->fl_type & LOCK_MAND) {
2028 out += sprintf(out, "%s ",
2029 (fl->fl_type & LOCK_READ)
2030 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2031 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2033 out += sprintf(out, "%s ",
2034 (fl->fl_type & F_INPROGRESS)
2035 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2036 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2039 #ifdef WE_CAN_BREAK_LSLK_NOW
2040 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2041 inode->i_sb->s_id, inode->i_ino);
2043 /* userspace relies on this representation of dev_t ;-( */
2044 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2045 MAJOR(inode->i_sb->s_dev),
2046 MINOR(inode->i_sb->s_dev), inode->i_ino);
2049 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2052 if (fl->fl_end == OFFSET_MAX)
2053 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2055 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2058 out += sprintf(out, "0 EOF\n");
2062 static void move_lock_status(char **p, off_t* pos, off_t offset)
2066 if(*pos >= offset) {
2067 /* the complete line is valid */
2072 if(*pos+len > offset) {
2073 /* use the second part of the line */
2074 int i = offset-*pos;
2075 memmove(*p,*p+i,len-i);
2080 /* discard the complete line */
2085 * get_locks_status - reports lock usage in /proc/locks
2086 * @buffer: address in userspace to write into
2088 * @offset: how far we are through the buffer
2089 * @length: how much to read
2092 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2094 struct list_head *tmp;
2100 list_for_each(tmp, &file_lock_list) {
2101 struct list_head *btmp;
2102 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2103 lock_get_status(q, fl, ++i, "");
2104 move_lock_status(&q, &pos, offset);
2106 if(pos >= offset+length)
2109 list_for_each(btmp, &fl->fl_block) {
2110 struct file_lock *bfl = list_entry(btmp,
2111 struct file_lock, fl_block);
2112 lock_get_status(q, bfl, i, " ->");
2113 move_lock_status(&q, &pos, offset);
2115 if(pos >= offset+length)
2122 if(q-buffer < length)
2128 * lock_may_read - checks that the region is free of locks
2129 * @inode: the inode that is being read
2130 * @start: the first byte to read
2131 * @len: the number of bytes to read
2133 * Emulates Windows locking requirements. Whole-file
2134 * mandatory locks (share modes) can prohibit a read and
2135 * byte-range POSIX locks can prohibit a read if they overlap.
2137 * N.B. this function is only ever called
2138 * from knfsd and ownership of locks is never checked.
2140 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2142 struct file_lock *fl;
2145 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2147 if (fl->fl_type == F_RDLCK)
2149 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2151 } else if (IS_FLOCK(fl)) {
2152 if (!(fl->fl_type & LOCK_MAND))
2154 if (fl->fl_type & LOCK_READ)
2165 EXPORT_SYMBOL(lock_may_read);
2168 * lock_may_write - checks that the region is free of locks
2169 * @inode: the inode that is being written
2170 * @start: the first byte to write
2171 * @len: the number of bytes to write
2173 * Emulates Windows locking requirements. Whole-file
2174 * mandatory locks (share modes) can prohibit a write and
2175 * byte-range POSIX locks can prohibit a write if they overlap.
2177 * N.B. this function is only ever called
2178 * from knfsd and ownership of locks is never checked.
2180 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2182 struct file_lock *fl;
2185 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2187 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2189 } else if (IS_FLOCK(fl)) {
2190 if (!(fl->fl_type & LOCK_MAND))
2192 if (fl->fl_type & LOCK_WRITE)
2203 EXPORT_SYMBOL(lock_may_write);
2205 static inline void __steal_locks(struct file *file, fl_owner_t from)
2207 struct inode *inode = file->f_dentry->d_inode;
2208 struct file_lock *fl = inode->i_flock;
2211 if (fl->fl_file == file && fl->fl_owner == from)
2212 fl->fl_owner = current->files;
2217 /* When getting ready for executing a binary, we make sure that current
2218 * has a files_struct on its own. Before dropping the old files_struct,
2219 * we take over ownership of all locks for all file descriptors we own.
2220 * Note that we may accidentally steal a lock for a file that a sibling
2221 * has created since the unshare_files() call.
2223 void steal_locks(fl_owner_t from)
2225 struct files_struct *files = current->files;
2227 struct fdtable *fdt;
2235 fdt = files_fdtable(files);
2239 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2241 set = fdt->open_fds->fds_bits[j++];
2244 struct file *file = fdt->fd[i];
2246 __steal_locks(file, from);
2255 EXPORT_SYMBOL(steal_locks);
2257 static int __init filelock_init(void)
2259 filelock_cache = kmem_cache_create("file_lock_cache",
2260 sizeof(struct file_lock), 0, SLAB_PANIC,
2265 core_initcall(filelock_init);