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 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
145 static kmem_cache_t *filelock_cache __read_mostly;
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
150 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
153 static void locks_release_private(struct file_lock *fl)
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
175 locks_release_private(fl);
176 kmem_cache_free(filelock_cache, fl);
179 void locks_init_lock(struct file_lock *fl)
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
185 fl->fl_fasync = NULL;
191 fl->fl_start = fl->fl_end = 0;
196 EXPORT_SYMBOL(locks_init_lock);
199 * Initialises the fields of the file lock which are invariant for
202 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
204 struct file_lock *lock = (struct file_lock *) foo;
206 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
207 SLAB_CTOR_CONSTRUCTOR)
210 locks_init_lock(lock);
213 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
216 if (fl->fl_ops->fl_copy_lock)
217 fl->fl_ops->fl_copy_lock(new, fl);
218 new->fl_ops = fl->fl_ops;
221 if (fl->fl_lmops->fl_copy_lock)
222 fl->fl_lmops->fl_copy_lock(new, fl);
223 new->fl_lmops = fl->fl_lmops;
228 * Initialize a new lock from an existing file_lock structure.
230 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
232 new->fl_owner = fl->fl_owner;
233 new->fl_pid = fl->fl_pid;
235 new->fl_flags = fl->fl_flags;
236 new->fl_type = fl->fl_type;
237 new->fl_start = fl->fl_start;
238 new->fl_end = fl->fl_end;
240 new->fl_lmops = NULL;
243 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
245 locks_release_private(new);
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
252 locks_copy_private(new, fl);
255 EXPORT_SYMBOL(locks_copy_lock);
257 static inline int flock_translate_cmd(int cmd) {
259 return cmd & (LOCK_MAND | LOCK_RW);
271 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
272 static int flock_make_lock(struct file *filp, struct file_lock **lock,
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
280 fl = locks_alloc_lock();
285 fl->fl_pid = current->tgid;
286 fl->fl_flags = FL_FLOCK;
288 fl->fl_end = OFFSET_MAX;
294 static int assign_type(struct file_lock *fl, int type)
308 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
311 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
316 switch (l->l_whence) {
324 start = i_size_read(filp->f_dentry->d_inode);
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
335 fl->fl_end = OFFSET_MAX;
337 end = start + l->l_len - 1;
339 } else if (l->l_len < 0) {
346 fl->fl_start = start; /* we record the absolute position */
347 if (fl->fl_end < fl->fl_start)
350 fl->fl_owner = current->files;
351 fl->fl_pid = current->tgid;
353 fl->fl_flags = FL_POSIX;
357 return assign_type(fl, l->l_type);
360 #if BITS_PER_LONG == 32
361 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
366 switch (l->l_whence) {
374 start = i_size_read(filp->f_dentry->d_inode);
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_end = start + l->l_len - 1;
386 } else if (l->l_len < 0) {
387 fl->fl_end = start - 1;
392 fl->fl_start = start; /* we record the absolute position */
393 if (fl->fl_end < fl->fl_start)
396 fl->fl_owner = current->files;
397 fl->fl_pid = current->tgid;
399 fl->fl_flags = FL_POSIX;
407 fl->fl_type = l->l_type;
417 /* default lease lock manager operations */
418 static void lease_break_callback(struct file_lock *fl)
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
423 static void lease_release_private_callback(struct file_lock *fl)
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
432 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
434 return fl->fl_file == try->fl_file;
437 static struct lock_manager_operations lease_manager_ops = {
438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
445 * Initialize a lease, use the default lock manager operations
447 static int lease_init(struct file *filp, int type, struct file_lock *fl)
449 fl->fl_owner = current->files;
450 fl->fl_pid = current->tgid;
453 fl->fl_flags = FL_LEASE;
454 if (assign_type(fl, type) != 0) {
459 fl->fl_end = OFFSET_MAX;
461 fl->fl_lmops = &lease_manager_ops;
465 /* Allocate a file_lock initialised to this type of lease */
466 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
468 struct file_lock *fl = locks_alloc_lock();
474 error = lease_init(filp, type, fl);
481 /* Check if two locks overlap each other.
483 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485 return ((fl1->fl_end >= fl2->fl_start) &&
486 (fl2->fl_end >= fl1->fl_start));
490 * Check whether two locks have the same owner.
492 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
494 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
495 return fl2->fl_lmops == fl1->fl_lmops &&
496 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
497 return fl1->fl_owner == fl2->fl_owner;
500 /* Remove waiter from blocker's block list.
501 * When blocker ends up pointing to itself then the list is empty.
503 static void __locks_delete_block(struct file_lock *waiter)
505 list_del_init(&waiter->fl_block);
506 list_del_init(&waiter->fl_link);
507 waiter->fl_next = NULL;
512 static void locks_delete_block(struct file_lock *waiter)
515 __locks_delete_block(waiter);
519 /* Insert waiter into blocker's block list.
520 * We use a circular list so that processes can be easily woken up in
521 * the order they blocked. The documentation doesn't require this but
522 * it seems like the reasonable thing to do.
524 static void locks_insert_block(struct file_lock *blocker,
525 struct file_lock *waiter)
527 BUG_ON(!list_empty(&waiter->fl_block));
528 list_add_tail(&waiter->fl_block, &blocker->fl_block);
529 waiter->fl_next = blocker;
530 if (IS_POSIX(blocker))
531 list_add(&waiter->fl_link, &blocked_list);
534 /* Wake up processes blocked waiting for blocker.
535 * If told to wait then schedule the processes until the block list
536 * is empty, otherwise empty the block list ourselves.
538 static void locks_wake_up_blocks(struct file_lock *blocker)
540 while (!list_empty(&blocker->fl_block)) {
541 struct file_lock *waiter = list_entry(blocker->fl_block.next,
542 struct file_lock, fl_block);
543 __locks_delete_block(waiter);
544 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
545 waiter->fl_lmops->fl_notify(waiter);
547 wake_up(&waiter->fl_wait);
551 /* Insert file lock fl into an inode's lock list at the position indicated
552 * by pos. At the same time add the lock to the global file lock list.
554 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
556 list_add(&fl->fl_link, &file_lock_list);
558 /* insert into file's list */
562 if (fl->fl_ops && fl->fl_ops->fl_insert)
563 fl->fl_ops->fl_insert(fl);
567 * Delete a lock and then free it.
568 * Wake up processes that are blocked waiting for this lock,
569 * notify the FS that the lock has been cleared and
570 * finally free the lock.
572 static void locks_delete_lock(struct file_lock **thisfl_p)
574 struct file_lock *fl = *thisfl_p;
576 *thisfl_p = fl->fl_next;
578 list_del_init(&fl->fl_link);
580 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
581 if (fl->fl_fasync != NULL) {
582 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
583 fl->fl_fasync = NULL;
586 if (fl->fl_ops && fl->fl_ops->fl_remove)
587 fl->fl_ops->fl_remove(fl);
589 locks_wake_up_blocks(fl);
593 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
594 * checks for shared/exclusive status of overlapping locks.
596 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
598 if (sys_fl->fl_type == F_WRLCK)
600 if (caller_fl->fl_type == F_WRLCK)
605 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
606 * checking before calling the locks_conflict().
608 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
610 /* POSIX locks owned by the same process do not conflict with
613 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
616 /* Check whether they overlap */
617 if (!locks_overlap(caller_fl, sys_fl))
620 return (locks_conflict(caller_fl, sys_fl));
623 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
624 * checking before calling the locks_conflict().
626 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
628 /* FLOCK locks referring to the same filp do not conflict with
631 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
633 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
636 return (locks_conflict(caller_fl, sys_fl));
639 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
642 DECLARE_WAITQUEUE(wait, current);
644 __set_current_state(TASK_INTERRUPTIBLE);
645 add_wait_queue(fl_wait, &wait);
649 result = schedule_timeout(timeout);
650 if (signal_pending(current))
651 result = -ERESTARTSYS;
652 remove_wait_queue(fl_wait, &wait);
653 __set_current_state(TASK_RUNNING);
657 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
660 locks_insert_block(blocker, waiter);
661 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
662 __locks_delete_block(waiter);
667 posix_test_lock(struct file *filp, struct file_lock *fl,
668 struct file_lock *conflock)
670 struct file_lock *cfl;
673 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
676 if (posix_locks_conflict(cfl, fl))
680 __locks_copy_lock(conflock, cfl);
688 EXPORT_SYMBOL(posix_test_lock);
690 /* This function tests for deadlock condition before putting a process to
691 * sleep. The detection scheme is no longer recursive. Recursive was neat,
692 * but dangerous - we risked stack corruption if the lock data was bad, or
693 * if the recursion was too deep for any other reason.
695 * We rely on the fact that a task can only be on one lock's wait queue
696 * at a time. When we find blocked_task on a wait queue we can re-search
697 * with blocked_task equal to that queue's owner, until either blocked_task
698 * isn't found, or blocked_task is found on a queue owned by my_task.
700 * Note: the above assumption may not be true when handling lock requests
701 * from a broken NFS client. But broken NFS clients have a lot more to
702 * worry about than proper deadlock detection anyway... --okir
704 int posix_locks_deadlock(struct file_lock *caller_fl,
705 struct file_lock *block_fl)
707 struct list_head *tmp;
710 if (posix_same_owner(caller_fl, block_fl))
712 list_for_each(tmp, &blocked_list) {
713 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
714 if (posix_same_owner(fl, block_fl)) {
723 EXPORT_SYMBOL(posix_locks_deadlock);
725 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
726 * at the head of the list, but that's secret knowledge known only to
727 * flock_lock_file and posix_lock_file.
729 static int flock_lock_file(struct file *filp, struct file_lock *request)
731 struct file_lock *new_fl = NULL;
732 struct file_lock **before;
733 struct inode * inode = filp->f_dentry->d_inode;
738 for_each_lock(inode, before) {
739 struct file_lock *fl = *before;
744 if (filp != fl->fl_file)
746 if (request->fl_type == fl->fl_type)
749 locks_delete_lock(before);
753 if (request->fl_type == F_UNLCK)
756 new_fl = locks_alloc_lock();
760 * If a higher-priority process was blocked on the old file lock,
761 * give it the opportunity to lock the file.
766 for_each_lock(inode, before) {
767 struct file_lock *fl = *before;
772 if (!flock_locks_conflict(request, fl))
775 if (request->fl_flags & FL_SLEEP)
776 locks_insert_block(fl, request);
779 locks_copy_lock(new_fl, request);
780 locks_insert_lock(&inode->i_flock, new_fl);
786 locks_free_lock(new_fl);
790 static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
792 struct file_lock *fl;
793 struct file_lock *new_fl, *new_fl2;
794 struct file_lock *left = NULL;
795 struct file_lock *right = NULL;
796 struct file_lock **before;
797 int error, added = 0;
800 * We may need two file_lock structures for this operation,
801 * so we get them in advance to avoid races.
803 new_fl = locks_alloc_lock();
804 new_fl2 = locks_alloc_lock();
807 if (request->fl_type != F_UNLCK) {
808 for_each_lock(inode, before) {
809 struct file_lock *fl = *before;
812 if (!posix_locks_conflict(request, fl))
815 locks_copy_lock(conflock, 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_conf(filp->f_dentry->d_inode, fl, NULL);
987 EXPORT_SYMBOL(posix_lock_file);
990 * posix_lock_file_conf - Apply a POSIX-style lock to a file
991 * @filp: The file to apply the lock to
992 * @fl: The lock to be applied
993 * @conflock: Place to return a copy of the conflicting lock, if found.
995 * Except for the conflock parameter, acts just like posix_lock_file.
997 int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
998 struct file_lock *conflock)
1000 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1002 EXPORT_SYMBOL(posix_lock_file_conf);
1005 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1006 * @filp: The file to apply the lock to
1007 * @fl: The lock to be applied
1009 * Add a POSIX style lock to a file.
1010 * We merge adjacent & overlapping locks whenever possible.
1011 * POSIX locks are sorted by owner task, then by starting address
1013 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1018 error = posix_lock_file(filp, fl);
1019 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1021 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1025 locks_delete_block(fl);
1030 EXPORT_SYMBOL(posix_lock_file_wait);
1033 * locks_mandatory_locked - Check for an active lock
1034 * @inode: the file to check
1036 * Searches the inode's list of locks to find any POSIX locks which conflict.
1037 * This function is called from locks_verify_locked() only.
1039 int locks_mandatory_locked(struct inode *inode)
1041 fl_owner_t owner = current->files;
1042 struct file_lock *fl;
1045 * Search the lock list for this inode for any POSIX locks.
1048 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1051 if (fl->fl_owner != owner)
1055 return fl ? -EAGAIN : 0;
1059 * locks_mandatory_area - Check for a conflicting lock
1060 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1062 * @inode: the file to check
1063 * @filp: how the file was opened (if it was)
1064 * @offset: start of area to check
1065 * @count: length of area to check
1067 * Searches the inode's list of locks to find any POSIX locks which conflict.
1068 * This function is called from rw_verify_area() and
1069 * locks_verify_truncate().
1071 int locks_mandatory_area(int read_write, struct inode *inode,
1072 struct file *filp, loff_t offset,
1075 struct file_lock fl;
1078 locks_init_lock(&fl);
1079 fl.fl_owner = current->files;
1080 fl.fl_pid = current->tgid;
1082 fl.fl_flags = FL_POSIX | FL_ACCESS;
1083 if (filp && !(filp->f_flags & O_NONBLOCK))
1084 fl.fl_flags |= FL_SLEEP;
1085 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1086 fl.fl_start = offset;
1087 fl.fl_end = offset + count - 1;
1090 error = __posix_lock_file_conf(inode, &fl, NULL);
1091 if (error != -EAGAIN)
1093 if (!(fl.fl_flags & FL_SLEEP))
1095 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1098 * If we've been sleeping someone might have
1099 * changed the permissions behind our back.
1101 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1105 locks_delete_block(&fl);
1112 EXPORT_SYMBOL(locks_mandatory_area);
1114 /* We already had a lease on this file; just change its type */
1115 int lease_modify(struct file_lock **before, int arg)
1117 struct file_lock *fl = *before;
1118 int error = assign_type(fl, arg);
1122 locks_wake_up_blocks(fl);
1124 locks_delete_lock(before);
1128 EXPORT_SYMBOL(lease_modify);
1130 static void time_out_leases(struct inode *inode)
1132 struct file_lock **before;
1133 struct file_lock *fl;
1135 before = &inode->i_flock;
1136 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1137 if ((fl->fl_break_time == 0)
1138 || time_before(jiffies, fl->fl_break_time)) {
1139 before = &fl->fl_next;
1142 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1143 if (fl == *before) /* lease_modify may have freed fl */
1144 before = &fl->fl_next;
1149 * __break_lease - revoke all outstanding leases on file
1150 * @inode: the inode of the file to return
1151 * @mode: the open mode (read or write)
1153 * break_lease (inlined for speed) has checked there already
1154 * is a lease on this file. Leases are broken on a call to open()
1155 * or truncate(). This function can sleep unless you
1156 * specified %O_NONBLOCK to your open().
1158 int __break_lease(struct inode *inode, unsigned int mode)
1160 int error = 0, future;
1161 struct file_lock *new_fl, *flock;
1162 struct file_lock *fl;
1164 unsigned long break_time;
1165 int i_have_this_lease = 0;
1167 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1172 time_out_leases(inode);
1174 flock = inode->i_flock;
1175 if ((flock == NULL) || !IS_LEASE(flock))
1178 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1179 if (fl->fl_owner == current->files)
1180 i_have_this_lease = 1;
1182 if (mode & FMODE_WRITE) {
1183 /* If we want write access, we have to revoke any lease. */
1184 future = F_UNLCK | F_INPROGRESS;
1185 } else if (flock->fl_type & F_INPROGRESS) {
1186 /* If the lease is already being broken, we just leave it */
1187 future = flock->fl_type;
1188 } else if (flock->fl_type & F_WRLCK) {
1189 /* Downgrade the exclusive lease to a read-only lease. */
1190 future = F_RDLCK | F_INPROGRESS;
1192 /* the existing lease was read-only, so we can read too. */
1196 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1202 if (lease_break_time > 0) {
1203 break_time = jiffies + lease_break_time * HZ;
1204 if (break_time == 0)
1205 break_time++; /* so that 0 means no break time */
1208 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1209 if (fl->fl_type != future) {
1210 fl->fl_type = future;
1211 fl->fl_break_time = break_time;
1212 /* lease must have lmops break callback */
1213 fl->fl_lmops->fl_break(fl);
1217 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1218 error = -EWOULDBLOCK;
1223 break_time = flock->fl_break_time;
1224 if (break_time != 0) {
1225 break_time -= jiffies;
1226 if (break_time == 0)
1229 error = locks_block_on_timeout(flock, new_fl, break_time);
1232 time_out_leases(inode);
1233 /* Wait for the next lease that has not been broken yet */
1234 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1235 flock = flock->fl_next) {
1236 if (flock->fl_type & F_INPROGRESS)
1245 locks_free_lock(new_fl);
1249 EXPORT_SYMBOL(__break_lease);
1254 * @time: pointer to a timespec which will contain the last modified time
1256 * This is to force NFS clients to flush their caches for files with
1257 * exclusive leases. The justification is that if someone has an
1258 * exclusive lease, then they could be modifiying it.
1260 void lease_get_mtime(struct inode *inode, struct timespec *time)
1262 struct file_lock *flock = inode->i_flock;
1263 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1264 *time = current_fs_time(inode->i_sb);
1266 *time = inode->i_mtime;
1269 EXPORT_SYMBOL(lease_get_mtime);
1272 * fcntl_getlease - Enquire what lease is currently active
1275 * The value returned by this function will be one of
1276 * (if no lease break is pending):
1278 * %F_RDLCK to indicate a shared lease is held.
1280 * %F_WRLCK to indicate an exclusive lease is held.
1282 * %F_UNLCK to indicate no lease is held.
1284 * (if a lease break is pending):
1286 * %F_RDLCK to indicate an exclusive lease needs to be
1287 * changed to a shared lease (or removed).
1289 * %F_UNLCK to indicate the lease needs to be removed.
1291 * XXX: sfr & willy disagree over whether F_INPROGRESS
1292 * should be returned to userspace.
1294 int fcntl_getlease(struct file *filp)
1296 struct file_lock *fl;
1300 time_out_leases(filp->f_dentry->d_inode);
1301 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1303 if (fl->fl_file == filp) {
1304 type = fl->fl_type & ~F_INPROGRESS;
1313 * __setlease - sets a lease on an open file
1314 * @filp: file pointer
1315 * @arg: type of lease to obtain
1316 * @flp: input - file_lock to use, output - file_lock inserted
1318 * The (input) flp->fl_lmops->fl_break function is required
1321 * Called with kernel lock held.
1323 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1325 struct file_lock *fl, **before, **my_before = NULL, *lease;
1326 struct dentry *dentry = filp->f_dentry;
1327 struct inode *inode = dentry->d_inode;
1328 int error, rdlease_count = 0, wrlease_count = 0;
1330 time_out_leases(inode);
1333 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1339 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1341 if ((arg == F_WRLCK)
1342 && ((atomic_read(&dentry->d_count) > 1)
1343 || (atomic_read(&inode->i_count) > 1)))
1347 * At this point, we know that if there is an exclusive
1348 * lease on this file, then we hold it on this filp
1349 * (otherwise our open of this file would have blocked).
1350 * And if we are trying to acquire an exclusive lease,
1351 * then the file is not open by anyone (including us)
1352 * except for this filp.
1354 for (before = &inode->i_flock;
1355 ((fl = *before) != NULL) && IS_LEASE(fl);
1356 before = &fl->fl_next) {
1357 if (lease->fl_lmops->fl_mylease(fl, lease))
1359 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1361 * Someone is in the process of opening this
1362 * file for writing so we may not take an
1363 * exclusive lease on it.
1370 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1371 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1374 if (my_before != NULL) {
1375 error = lease->fl_lmops->fl_change(my_before, arg);
1387 error = lease_alloc(filp, arg, &fl);
1391 locks_copy_lock(fl, lease);
1393 locks_insert_lock(before, fl);
1401 * setlease - sets a lease on an open file
1402 * @filp: file pointer
1403 * @arg: type of lease to obtain
1404 * @lease: file_lock to use
1406 * Call this to establish a lease on the file.
1407 * The fl_lmops fl_break function is required by break_lease
1410 int setlease(struct file *filp, long arg, struct file_lock **lease)
1412 struct dentry *dentry = filp->f_dentry;
1413 struct inode *inode = dentry->d_inode;
1416 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1418 if (!S_ISREG(inode->i_mode))
1420 error = security_file_lock(filp, arg);
1425 error = __setlease(filp, arg, lease);
1431 EXPORT_SYMBOL(setlease);
1434 * fcntl_setlease - sets a lease on an open file
1435 * @fd: open file descriptor
1436 * @filp: file pointer
1437 * @arg: type of lease to obtain
1439 * Call this fcntl to establish a lease on the file.
1440 * Note that you also need to call %F_SETSIG to
1441 * receive a signal when the lease is broken.
1443 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1445 struct file_lock fl, *flp = &fl;
1446 struct dentry *dentry = filp->f_dentry;
1447 struct inode *inode = dentry->d_inode;
1450 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1452 if (!S_ISREG(inode->i_mode))
1454 error = security_file_lock(filp, arg);
1458 locks_init_lock(&fl);
1459 error = lease_init(filp, arg, &fl);
1465 error = __setlease(filp, arg, &flp);
1466 if (error || arg == F_UNLCK)
1469 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1471 /* remove lease just inserted by __setlease */
1472 flp->fl_type = F_UNLCK | F_INPROGRESS;
1473 flp->fl_break_time = jiffies- 10;
1474 time_out_leases(inode);
1478 error = f_setown(filp, current->pid, 0);
1485 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1486 * @filp: The file to apply the lock to
1487 * @fl: The lock to be applied
1489 * Add a FLOCK style lock to a file.
1491 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1496 error = flock_lock_file(filp, fl);
1497 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1499 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1503 locks_delete_block(fl);
1509 EXPORT_SYMBOL(flock_lock_file_wait);
1512 * sys_flock: - flock() system call.
1513 * @fd: the file descriptor to lock.
1514 * @cmd: the type of lock to apply.
1516 * Apply a %FL_FLOCK style lock to an open file descriptor.
1517 * The @cmd can be one of
1519 * %LOCK_SH -- a shared lock.
1521 * %LOCK_EX -- an exclusive lock.
1523 * %LOCK_UN -- remove an existing lock.
1525 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1527 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1528 * processes read and write access respectively.
1530 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1533 struct file_lock *lock;
1534 int can_sleep, unlock;
1542 can_sleep = !(cmd & LOCK_NB);
1544 unlock = (cmd == LOCK_UN);
1546 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1549 error = flock_make_lock(filp, &lock, cmd);
1553 lock->fl_flags |= FL_SLEEP;
1555 error = security_file_lock(filp, cmd);
1559 if (filp->f_op && filp->f_op->flock)
1560 error = filp->f_op->flock(filp,
1561 (can_sleep) ? F_SETLKW : F_SETLK,
1564 error = flock_lock_file_wait(filp, lock);
1567 locks_free_lock(lock);
1575 /* Report the first existing lock that would conflict with l.
1576 * This implements the F_GETLK command of fcntl().
1578 int fcntl_getlk(struct file *filp, struct flock __user *l)
1580 struct file_lock *fl, cfl, file_lock;
1585 if (copy_from_user(&flock, l, sizeof(flock)))
1588 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1591 error = flock_to_posix_lock(filp, &file_lock, &flock);
1595 if (filp->f_op && filp->f_op->lock) {
1596 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1597 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1598 file_lock.fl_ops->fl_release_private(&file_lock);
1602 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1604 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1607 flock.l_type = F_UNLCK;
1609 flock.l_pid = fl->fl_pid;
1610 #if BITS_PER_LONG == 32
1612 * Make sure we can represent the posix lock via
1613 * legacy 32bit flock.
1616 if (fl->fl_start > OFFT_OFFSET_MAX)
1618 if ((fl->fl_end != OFFSET_MAX)
1619 && (fl->fl_end > OFFT_OFFSET_MAX))
1622 flock.l_start = fl->fl_start;
1623 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1624 fl->fl_end - fl->fl_start + 1;
1626 flock.l_type = fl->fl_type;
1629 if (!copy_to_user(l, &flock, sizeof(flock)))
1635 /* Apply the lock described by l to an open file descriptor.
1636 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1638 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1639 struct flock __user *l)
1641 struct file_lock *file_lock = locks_alloc_lock();
1643 struct inode *inode;
1646 if (file_lock == NULL)
1650 * This might block, so we do it before checking the inode.
1653 if (copy_from_user(&flock, l, sizeof(flock)))
1656 inode = filp->f_dentry->d_inode;
1658 /* Don't allow mandatory locks on files that may be memory mapped
1661 if (IS_MANDLOCK(inode) &&
1662 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1663 mapping_writably_mapped(filp->f_mapping)) {
1669 error = flock_to_posix_lock(filp, file_lock, &flock);
1672 if (cmd == F_SETLKW) {
1673 file_lock->fl_flags |= FL_SLEEP;
1677 switch (flock.l_type) {
1679 if (!(filp->f_mode & FMODE_READ))
1683 if (!(filp->f_mode & FMODE_WRITE))
1693 error = security_file_lock(filp, file_lock->fl_type);
1697 if (filp->f_op && filp->f_op->lock != NULL)
1698 error = filp->f_op->lock(filp, cmd, file_lock);
1701 error = posix_lock_file(filp, file_lock);
1702 if ((error != -EAGAIN) || (cmd == F_SETLK))
1704 error = wait_event_interruptible(file_lock->fl_wait,
1705 !file_lock->fl_next);
1709 locks_delete_block(file_lock);
1715 * Attempt to detect a close/fcntl race and recover by
1716 * releasing the lock that was just acquired.
1718 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1719 flock.l_type = F_UNLCK;
1724 locks_free_lock(file_lock);
1728 #if BITS_PER_LONG == 32
1729 /* Report the first existing lock that would conflict with l.
1730 * This implements the F_GETLK command of fcntl().
1732 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1734 struct file_lock *fl, cfl, file_lock;
1735 struct flock64 flock;
1739 if (copy_from_user(&flock, l, sizeof(flock)))
1742 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1745 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1749 if (filp->f_op && filp->f_op->lock) {
1750 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1751 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1752 file_lock.fl_ops->fl_release_private(&file_lock);
1756 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1758 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1761 flock.l_type = F_UNLCK;
1763 flock.l_pid = fl->fl_pid;
1764 flock.l_start = fl->fl_start;
1765 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1766 fl->fl_end - fl->fl_start + 1;
1768 flock.l_type = fl->fl_type;
1771 if (!copy_to_user(l, &flock, sizeof(flock)))
1778 /* Apply the lock described by l to an open file descriptor.
1779 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1781 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1782 struct flock64 __user *l)
1784 struct file_lock *file_lock = locks_alloc_lock();
1785 struct flock64 flock;
1786 struct inode *inode;
1789 if (file_lock == NULL)
1793 * This might block, so we do it before checking the inode.
1796 if (copy_from_user(&flock, l, sizeof(flock)))
1799 inode = filp->f_dentry->d_inode;
1801 /* Don't allow mandatory locks on files that may be memory mapped
1804 if (IS_MANDLOCK(inode) &&
1805 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1806 mapping_writably_mapped(filp->f_mapping)) {
1812 error = flock64_to_posix_lock(filp, file_lock, &flock);
1815 if (cmd == F_SETLKW64) {
1816 file_lock->fl_flags |= FL_SLEEP;
1820 switch (flock.l_type) {
1822 if (!(filp->f_mode & FMODE_READ))
1826 if (!(filp->f_mode & FMODE_WRITE))
1836 error = security_file_lock(filp, file_lock->fl_type);
1840 if (filp->f_op && filp->f_op->lock != NULL)
1841 error = filp->f_op->lock(filp, cmd, file_lock);
1844 error = posix_lock_file(filp, file_lock);
1845 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1847 error = wait_event_interruptible(file_lock->fl_wait,
1848 !file_lock->fl_next);
1852 locks_delete_block(file_lock);
1858 * Attempt to detect a close/fcntl race and recover by
1859 * releasing the lock that was just acquired.
1861 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1862 flock.l_type = F_UNLCK;
1867 locks_free_lock(file_lock);
1870 #endif /* BITS_PER_LONG == 32 */
1873 * This function is called when the file is being removed
1874 * from the task's fd array. POSIX locks belonging to this task
1875 * are deleted at this time.
1877 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1879 struct file_lock lock, **before;
1882 * If there are no locks held on this file, we don't need to call
1883 * posix_lock_file(). Another process could be setting a lock on this
1884 * file at the same time, but we wouldn't remove that lock anyway.
1886 before = &filp->f_dentry->d_inode->i_flock;
1887 if (*before == NULL)
1890 lock.fl_type = F_UNLCK;
1891 lock.fl_flags = FL_POSIX;
1893 lock.fl_end = OFFSET_MAX;
1894 lock.fl_owner = owner;
1895 lock.fl_pid = current->tgid;
1896 lock.fl_file = filp;
1898 lock.fl_lmops = NULL;
1900 if (filp->f_op && filp->f_op->lock != NULL) {
1901 filp->f_op->lock(filp, F_SETLK, &lock);
1905 /* Can't use posix_lock_file here; we need to remove it no matter
1906 * which pid we have.
1909 while (*before != NULL) {
1910 struct file_lock *fl = *before;
1911 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1912 locks_delete_lock(before);
1915 before = &fl->fl_next;
1919 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1920 lock.fl_ops->fl_release_private(&lock);
1923 EXPORT_SYMBOL(locks_remove_posix);
1926 * This function is called on the last close of an open file.
1928 void locks_remove_flock(struct file *filp)
1930 struct inode * inode = filp->f_dentry->d_inode;
1931 struct file_lock *fl;
1932 struct file_lock **before;
1934 if (!inode->i_flock)
1937 if (filp->f_op && filp->f_op->flock) {
1938 struct file_lock fl = {
1939 .fl_pid = current->tgid,
1941 .fl_flags = FL_FLOCK,
1943 .fl_end = OFFSET_MAX,
1945 filp->f_op->flock(filp, F_SETLKW, &fl);
1946 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1947 fl.fl_ops->fl_release_private(&fl);
1951 before = &inode->i_flock;
1953 while ((fl = *before) != NULL) {
1954 if (fl->fl_file == filp) {
1956 locks_delete_lock(before);
1960 lease_modify(before, F_UNLCK);
1966 before = &fl->fl_next;
1972 * posix_unblock_lock - stop waiting for a file lock
1973 * @filp: how the file was opened
1974 * @waiter: the lock which was waiting
1976 * lockd needs to block waiting for locks.
1979 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1984 if (waiter->fl_next)
1985 __locks_delete_block(waiter);
1992 EXPORT_SYMBOL(posix_unblock_lock);
1994 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1996 struct inode *inode = NULL;
1998 if (fl->fl_file != NULL)
1999 inode = fl->fl_file->f_dentry->d_inode;
2001 out += sprintf(out, "%d:%s ", id, pfx);
2003 out += sprintf(out, "%6s %s ",
2004 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2005 (inode == NULL) ? "*NOINODE*" :
2006 (IS_MANDLOCK(inode) &&
2007 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2008 "MANDATORY" : "ADVISORY ");
2009 } else if (IS_FLOCK(fl)) {
2010 if (fl->fl_type & LOCK_MAND) {
2011 out += sprintf(out, "FLOCK MSNFS ");
2013 out += sprintf(out, "FLOCK ADVISORY ");
2015 } else if (IS_LEASE(fl)) {
2016 out += sprintf(out, "LEASE ");
2017 if (fl->fl_type & F_INPROGRESS)
2018 out += sprintf(out, "BREAKING ");
2019 else if (fl->fl_file)
2020 out += sprintf(out, "ACTIVE ");
2022 out += sprintf(out, "BREAKER ");
2024 out += sprintf(out, "UNKNOWN UNKNOWN ");
2026 if (fl->fl_type & LOCK_MAND) {
2027 out += sprintf(out, "%s ",
2028 (fl->fl_type & LOCK_READ)
2029 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2030 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2032 out += sprintf(out, "%s ",
2033 (fl->fl_type & F_INPROGRESS)
2034 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2035 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2038 #ifdef WE_CAN_BREAK_LSLK_NOW
2039 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2040 inode->i_sb->s_id, inode->i_ino);
2042 /* userspace relies on this representation of dev_t ;-( */
2043 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2044 MAJOR(inode->i_sb->s_dev),
2045 MINOR(inode->i_sb->s_dev), inode->i_ino);
2048 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2051 if (fl->fl_end == OFFSET_MAX)
2052 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2054 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2057 out += sprintf(out, "0 EOF\n");
2061 static void move_lock_status(char **p, off_t* pos, off_t offset)
2065 if(*pos >= offset) {
2066 /* the complete line is valid */
2071 if(*pos+len > offset) {
2072 /* use the second part of the line */
2073 int i = offset-*pos;
2074 memmove(*p,*p+i,len-i);
2079 /* discard the complete line */
2084 * get_locks_status - reports lock usage in /proc/locks
2085 * @buffer: address in userspace to write into
2087 * @offset: how far we are through the buffer
2088 * @length: how much to read
2091 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2093 struct list_head *tmp;
2099 list_for_each(tmp, &file_lock_list) {
2100 struct list_head *btmp;
2101 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2102 lock_get_status(q, fl, ++i, "");
2103 move_lock_status(&q, &pos, offset);
2105 if(pos >= offset+length)
2108 list_for_each(btmp, &fl->fl_block) {
2109 struct file_lock *bfl = list_entry(btmp,
2110 struct file_lock, fl_block);
2111 lock_get_status(q, bfl, i, " ->");
2112 move_lock_status(&q, &pos, offset);
2114 if(pos >= offset+length)
2121 if(q-buffer < length)
2127 * lock_may_read - checks that the region is free of locks
2128 * @inode: the inode that is being read
2129 * @start: the first byte to read
2130 * @len: the number of bytes to read
2132 * Emulates Windows locking requirements. Whole-file
2133 * mandatory locks (share modes) can prohibit a read and
2134 * byte-range POSIX locks can prohibit a read if they overlap.
2136 * N.B. this function is only ever called
2137 * from knfsd and ownership of locks is never checked.
2139 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2141 struct file_lock *fl;
2144 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2146 if (fl->fl_type == F_RDLCK)
2148 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2150 } else if (IS_FLOCK(fl)) {
2151 if (!(fl->fl_type & LOCK_MAND))
2153 if (fl->fl_type & LOCK_READ)
2164 EXPORT_SYMBOL(lock_may_read);
2167 * lock_may_write - checks that the region is free of locks
2168 * @inode: the inode that is being written
2169 * @start: the first byte to write
2170 * @len: the number of bytes to write
2172 * Emulates Windows locking requirements. Whole-file
2173 * mandatory locks (share modes) can prohibit a write and
2174 * byte-range POSIX locks can prohibit a write if they overlap.
2176 * N.B. this function is only ever called
2177 * from knfsd and ownership of locks is never checked.
2179 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2181 struct file_lock *fl;
2184 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2186 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2188 } else if (IS_FLOCK(fl)) {
2189 if (!(fl->fl_type & LOCK_MAND))
2191 if (fl->fl_type & LOCK_WRITE)
2202 EXPORT_SYMBOL(lock_may_write);
2204 static inline void __steal_locks(struct file *file, fl_owner_t from)
2206 struct inode *inode = file->f_dentry->d_inode;
2207 struct file_lock *fl = inode->i_flock;
2210 if (fl->fl_file == file && fl->fl_owner == from)
2211 fl->fl_owner = current->files;
2216 /* When getting ready for executing a binary, we make sure that current
2217 * has a files_struct on its own. Before dropping the old files_struct,
2218 * we take over ownership of all locks for all file descriptors we own.
2219 * Note that we may accidentally steal a lock for a file that a sibling
2220 * has created since the unshare_files() call.
2222 void steal_locks(fl_owner_t from)
2224 struct files_struct *files = current->files;
2226 struct fdtable *fdt;
2234 fdt = files_fdtable(files);
2238 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2240 set = fdt->open_fds->fds_bits[j++];
2243 struct file *file = fdt->fd[i];
2245 __steal_locks(file, from);
2254 EXPORT_SYMBOL(steal_locks);
2256 static int __init filelock_init(void)
2258 filelock_cache = kmem_cache_create("file_lock_cache",
2259 sizeof(struct file_lock), 0, SLAB_PANIC,
2264 core_initcall(filelock_init);