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/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/module.h>
123 #include <linux/security.h>
124 #include <linux/slab.h>
125 #include <linux/smp_lock.h>
126 #include <linux/syscalls.h>
127 #include <linux/time.h>
128 #include <linux/rcupdate.h>
129 #include <linux/pid_namespace.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable = 1;
138 int lease_break_time = 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 static LIST_HEAD(file_lock_list);
144 static LIST_HEAD(blocked_list);
145 static DEFINE_SPINLOCK(file_lock_lock);
148 * Protects the two list heads above, plus the inode->i_flock list
149 * FIXME: should use a spinlock, once lockd and ceph are ready.
151 void lock_flocks(void)
153 spin_lock(&file_lock_lock);
155 EXPORT_SYMBOL_GPL(lock_flocks);
157 void unlock_flocks(void)
159 spin_unlock(&file_lock_lock);
161 EXPORT_SYMBOL_GPL(unlock_flocks);
163 static struct kmem_cache *filelock_cache __read_mostly;
165 /* Allocate an empty lock structure. */
166 struct file_lock *locks_alloc_lock(void)
168 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
170 EXPORT_SYMBOL_GPL(locks_alloc_lock);
172 void locks_release_private(struct file_lock *fl)
175 if (fl->fl_ops->fl_release_private)
176 fl->fl_ops->fl_release_private(fl);
180 if (fl->fl_lmops->fl_release_private)
181 fl->fl_lmops->fl_release_private(fl);
186 EXPORT_SYMBOL_GPL(locks_release_private);
188 /* Free a lock which is not in use. */
189 static void locks_free_lock(struct file_lock *fl)
191 BUG_ON(waitqueue_active(&fl->fl_wait));
192 BUG_ON(!list_empty(&fl->fl_block));
193 BUG_ON(!list_empty(&fl->fl_link));
195 locks_release_private(fl);
196 kmem_cache_free(filelock_cache, fl);
199 void locks_init_lock(struct file_lock *fl)
201 INIT_LIST_HEAD(&fl->fl_link);
202 INIT_LIST_HEAD(&fl->fl_block);
203 init_waitqueue_head(&fl->fl_wait);
205 fl->fl_fasync = NULL;
212 fl->fl_start = fl->fl_end = 0;
217 EXPORT_SYMBOL(locks_init_lock);
220 * Initialises the fields of the file lock which are invariant for
223 static void init_once(void *foo)
225 struct file_lock *lock = (struct file_lock *) foo;
227 locks_init_lock(lock);
230 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
233 if (fl->fl_ops->fl_copy_lock)
234 fl->fl_ops->fl_copy_lock(new, fl);
235 new->fl_ops = fl->fl_ops;
238 if (fl->fl_lmops->fl_copy_lock)
239 fl->fl_lmops->fl_copy_lock(new, fl);
240 new->fl_lmops = fl->fl_lmops;
245 * Initialize a new lock from an existing file_lock structure.
247 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
249 new->fl_owner = fl->fl_owner;
250 new->fl_pid = fl->fl_pid;
252 new->fl_flags = fl->fl_flags;
253 new->fl_type = fl->fl_type;
254 new->fl_start = fl->fl_start;
255 new->fl_end = fl->fl_end;
257 new->fl_lmops = NULL;
259 EXPORT_SYMBOL(__locks_copy_lock);
261 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
263 locks_release_private(new);
265 __locks_copy_lock(new, fl);
266 new->fl_file = fl->fl_file;
267 new->fl_ops = fl->fl_ops;
268 new->fl_lmops = fl->fl_lmops;
270 locks_copy_private(new, fl);
273 EXPORT_SYMBOL(locks_copy_lock);
275 static inline int flock_translate_cmd(int cmd) {
277 return cmd & (LOCK_MAND | LOCK_RW);
289 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
290 static int flock_make_lock(struct file *filp, struct file_lock **lock,
293 struct file_lock *fl;
294 int type = flock_translate_cmd(cmd);
298 fl = locks_alloc_lock();
303 fl->fl_pid = current->tgid;
304 fl->fl_flags = FL_FLOCK;
306 fl->fl_end = OFFSET_MAX;
312 static int assign_type(struct file_lock *fl, int type)
326 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
329 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
334 switch (l->l_whence) {
342 start = i_size_read(filp->f_path.dentry->d_inode);
348 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
349 POSIX-2001 defines it. */
353 fl->fl_end = OFFSET_MAX;
355 end = start + l->l_len - 1;
357 } else if (l->l_len < 0) {
364 fl->fl_start = start; /* we record the absolute position */
365 if (fl->fl_end < fl->fl_start)
368 fl->fl_owner = current->files;
369 fl->fl_pid = current->tgid;
371 fl->fl_flags = FL_POSIX;
375 return assign_type(fl, l->l_type);
378 #if BITS_PER_LONG == 32
379 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
384 switch (l->l_whence) {
392 start = i_size_read(filp->f_path.dentry->d_inode);
401 fl->fl_end = OFFSET_MAX;
403 fl->fl_end = start + l->l_len - 1;
404 } else if (l->l_len < 0) {
405 fl->fl_end = start - 1;
410 fl->fl_start = start; /* we record the absolute position */
411 if (fl->fl_end < fl->fl_start)
414 fl->fl_owner = current->files;
415 fl->fl_pid = current->tgid;
417 fl->fl_flags = FL_POSIX;
425 fl->fl_type = l->l_type;
435 /* default lease lock manager operations */
436 static void lease_break_callback(struct file_lock *fl)
438 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
441 static void lease_release_private_callback(struct file_lock *fl)
446 f_delown(fl->fl_file);
447 fl->fl_file->f_owner.signum = 0;
450 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
452 return fl->fl_file == try->fl_file;
455 static const struct lock_manager_operations lease_manager_ops = {
456 .fl_break = lease_break_callback,
457 .fl_release_private = lease_release_private_callback,
458 .fl_mylease = lease_mylease_callback,
459 .fl_change = lease_modify,
463 * Initialize a lease, use the default lock manager operations
465 static int lease_init(struct file *filp, int type, struct file_lock *fl)
467 if (assign_type(fl, type) != 0)
470 fl->fl_owner = current->files;
471 fl->fl_pid = current->tgid;
474 fl->fl_flags = FL_LEASE;
476 fl->fl_end = OFFSET_MAX;
478 fl->fl_lmops = &lease_manager_ops;
482 /* Allocate a file_lock initialised to this type of lease */
483 static struct file_lock *lease_alloc(struct file *filp, int type)
485 struct file_lock *fl = locks_alloc_lock();
489 return ERR_PTR(error);
491 error = lease_init(filp, type, fl);
494 return ERR_PTR(error);
499 /* Check if two locks overlap each other.
501 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
503 return ((fl1->fl_end >= fl2->fl_start) &&
504 (fl2->fl_end >= fl1->fl_start));
508 * Check whether two locks have the same owner.
510 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
512 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
513 return fl2->fl_lmops == fl1->fl_lmops &&
514 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
515 return fl1->fl_owner == fl2->fl_owner;
518 /* Remove waiter from blocker's block list.
519 * When blocker ends up pointing to itself then the list is empty.
521 static void __locks_delete_block(struct file_lock *waiter)
523 list_del_init(&waiter->fl_block);
524 list_del_init(&waiter->fl_link);
525 waiter->fl_next = NULL;
530 static void locks_delete_block(struct file_lock *waiter)
533 __locks_delete_block(waiter);
537 /* Insert waiter into blocker's block list.
538 * We use a circular list so that processes can be easily woken up in
539 * the order they blocked. The documentation doesn't require this but
540 * it seems like the reasonable thing to do.
542 static void locks_insert_block(struct file_lock *blocker,
543 struct file_lock *waiter)
545 BUG_ON(!list_empty(&waiter->fl_block));
546 list_add_tail(&waiter->fl_block, &blocker->fl_block);
547 waiter->fl_next = blocker;
548 if (IS_POSIX(blocker))
549 list_add(&waiter->fl_link, &blocked_list);
552 /* Wake up processes blocked waiting for blocker.
553 * If told to wait then schedule the processes until the block list
554 * is empty, otherwise empty the block list ourselves.
556 static void locks_wake_up_blocks(struct file_lock *blocker)
558 while (!list_empty(&blocker->fl_block)) {
559 struct file_lock *waiter;
561 waiter = list_first_entry(&blocker->fl_block,
562 struct file_lock, fl_block);
563 __locks_delete_block(waiter);
564 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
565 waiter->fl_lmops->fl_notify(waiter);
567 wake_up(&waiter->fl_wait);
571 /* Insert file lock fl into an inode's lock list at the position indicated
572 * by pos. At the same time add the lock to the global file lock list.
574 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
576 list_add(&fl->fl_link, &file_lock_list);
578 fl->fl_nspid = get_pid(task_tgid(current));
580 /* insert into file's list */
586 * Delete a lock and then free it.
587 * Wake up processes that are blocked waiting for this lock,
588 * notify the FS that the lock has been cleared and
589 * finally free the lock.
591 static void locks_delete_lock(struct file_lock **thisfl_p)
593 struct file_lock *fl = *thisfl_p;
595 *thisfl_p = fl->fl_next;
597 list_del_init(&fl->fl_link);
599 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
600 if (fl->fl_fasync != NULL) {
601 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
602 fl->fl_fasync = NULL;
606 put_pid(fl->fl_nspid);
610 locks_wake_up_blocks(fl);
614 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
615 * checks for shared/exclusive status of overlapping locks.
617 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
619 if (sys_fl->fl_type == F_WRLCK)
621 if (caller_fl->fl_type == F_WRLCK)
626 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
627 * checking before calling the locks_conflict().
629 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
631 /* POSIX locks owned by the same process do not conflict with
634 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
637 /* Check whether they overlap */
638 if (!locks_overlap(caller_fl, sys_fl))
641 return (locks_conflict(caller_fl, sys_fl));
644 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
645 * checking before calling the locks_conflict().
647 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
649 /* FLOCK locks referring to the same filp do not conflict with
652 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
654 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
657 return (locks_conflict(caller_fl, sys_fl));
661 posix_test_lock(struct file *filp, struct file_lock *fl)
663 struct file_lock *cfl;
666 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
669 if (posix_locks_conflict(fl, cfl))
673 __locks_copy_lock(fl, cfl);
675 fl->fl_pid = pid_vnr(cfl->fl_nspid);
677 fl->fl_type = F_UNLCK;
681 EXPORT_SYMBOL(posix_test_lock);
684 * Deadlock detection:
686 * We attempt to detect deadlocks that are due purely to posix file
689 * We assume that a task can be waiting for at most one lock at a time.
690 * So for any acquired lock, the process holding that lock may be
691 * waiting on at most one other lock. That lock in turns may be held by
692 * someone waiting for at most one other lock. Given a requested lock
693 * caller_fl which is about to wait for a conflicting lock block_fl, we
694 * follow this chain of waiters to ensure we are not about to create a
697 * Since we do this before we ever put a process to sleep on a lock, we
698 * are ensured that there is never a cycle; that is what guarantees that
699 * the while() loop in posix_locks_deadlock() eventually completes.
701 * Note: the above assumption may not be true when handling lock
702 * requests from a broken NFS client. It may also fail in the presence
703 * of tasks (such as posix threads) sharing the same open file table.
705 * To handle those cases, we just bail out after a few iterations.
708 #define MAX_DEADLK_ITERATIONS 10
710 /* Find a lock that the owner of the given block_fl is blocking on. */
711 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
713 struct file_lock *fl;
715 list_for_each_entry(fl, &blocked_list, fl_link) {
716 if (posix_same_owner(fl, block_fl))
722 static int posix_locks_deadlock(struct file_lock *caller_fl,
723 struct file_lock *block_fl)
727 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
728 if (i++ > MAX_DEADLK_ITERATIONS)
730 if (posix_same_owner(caller_fl, block_fl))
736 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
737 * after any leases, but before any posix locks.
739 * Note that if called with an FL_EXISTS argument, the caller may determine
740 * whether or not a lock was successfully freed by testing the return
743 static int flock_lock_file(struct file *filp, struct file_lock *request)
745 struct file_lock *new_fl = NULL;
746 struct file_lock **before;
747 struct inode * inode = filp->f_path.dentry->d_inode;
751 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
752 new_fl = locks_alloc_lock();
758 if (request->fl_flags & FL_ACCESS)
761 for_each_lock(inode, before) {
762 struct file_lock *fl = *before;
767 if (filp != fl->fl_file)
769 if (request->fl_type == fl->fl_type)
772 locks_delete_lock(before);
776 if (request->fl_type == F_UNLCK) {
777 if ((request->fl_flags & FL_EXISTS) && !found)
783 * If a higher-priority process was blocked on the old file lock,
784 * give it the opportunity to lock the file.
793 for_each_lock(inode, before) {
794 struct file_lock *fl = *before;
799 if (!flock_locks_conflict(request, fl))
802 if (!(request->fl_flags & FL_SLEEP))
804 error = FILE_LOCK_DEFERRED;
805 locks_insert_block(fl, request);
808 if (request->fl_flags & FL_ACCESS)
810 locks_copy_lock(new_fl, request);
811 locks_insert_lock(before, new_fl);
818 locks_free_lock(new_fl);
822 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
824 struct file_lock *fl;
825 struct file_lock *new_fl = NULL;
826 struct file_lock *new_fl2 = NULL;
827 struct file_lock *left = NULL;
828 struct file_lock *right = NULL;
829 struct file_lock **before;
830 int error, added = 0;
833 * We may need two file_lock structures for this operation,
834 * so we get them in advance to avoid races.
836 * In some cases we can be sure, that no new locks will be needed
838 if (!(request->fl_flags & FL_ACCESS) &&
839 (request->fl_type != F_UNLCK ||
840 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
841 new_fl = locks_alloc_lock();
842 new_fl2 = locks_alloc_lock();
846 if (request->fl_type != F_UNLCK) {
847 for_each_lock(inode, before) {
851 if (!posix_locks_conflict(request, fl))
854 __locks_copy_lock(conflock, fl);
856 if (!(request->fl_flags & FL_SLEEP))
859 if (posix_locks_deadlock(request, fl))
861 error = FILE_LOCK_DEFERRED;
862 locks_insert_block(fl, request);
867 /* If we're just looking for a conflict, we're done. */
869 if (request->fl_flags & FL_ACCESS)
873 * Find the first old lock with the same owner as the new lock.
876 before = &inode->i_flock;
878 /* First skip locks owned by other processes. */
879 while ((fl = *before) && (!IS_POSIX(fl) ||
880 !posix_same_owner(request, fl))) {
881 before = &fl->fl_next;
884 /* Process locks with this owner. */
885 while ((fl = *before) && posix_same_owner(request, fl)) {
886 /* Detect adjacent or overlapping regions (if same lock type)
888 if (request->fl_type == fl->fl_type) {
889 /* In all comparisons of start vs end, use
890 * "start - 1" rather than "end + 1". If end
891 * is OFFSET_MAX, end + 1 will become negative.
893 if (fl->fl_end < request->fl_start - 1)
895 /* If the next lock in the list has entirely bigger
896 * addresses than the new one, insert the lock here.
898 if (fl->fl_start - 1 > request->fl_end)
901 /* If we come here, the new and old lock are of the
902 * same type and adjacent or overlapping. Make one
903 * lock yielding from the lower start address of both
904 * locks to the higher end address.
906 if (fl->fl_start > request->fl_start)
907 fl->fl_start = request->fl_start;
909 request->fl_start = fl->fl_start;
910 if (fl->fl_end < request->fl_end)
911 fl->fl_end = request->fl_end;
913 request->fl_end = fl->fl_end;
915 locks_delete_lock(before);
922 /* Processing for different lock types is a bit
925 if (fl->fl_end < request->fl_start)
927 if (fl->fl_start > request->fl_end)
929 if (request->fl_type == F_UNLCK)
931 if (fl->fl_start < request->fl_start)
933 /* If the next lock in the list has a higher end
934 * address than the new one, insert the new one here.
936 if (fl->fl_end > request->fl_end) {
940 if (fl->fl_start >= request->fl_start) {
941 /* The new lock completely replaces an old
942 * one (This may happen several times).
945 locks_delete_lock(before);
948 /* Replace the old lock with the new one.
949 * Wake up anybody waiting for the old one,
950 * as the change in lock type might satisfy
953 locks_wake_up_blocks(fl);
954 fl->fl_start = request->fl_start;
955 fl->fl_end = request->fl_end;
956 fl->fl_type = request->fl_type;
957 locks_release_private(fl);
958 locks_copy_private(fl, request);
963 /* Go on to next lock.
966 before = &fl->fl_next;
970 * The above code only modifies existing locks in case of
971 * merging or replacing. If new lock(s) need to be inserted
972 * all modifications are done bellow this, so it's safe yet to
975 error = -ENOLCK; /* "no luck" */
976 if (right && left == right && !new_fl2)
981 if (request->fl_type == F_UNLCK) {
982 if (request->fl_flags & FL_EXISTS)
991 locks_copy_lock(new_fl, request);
992 locks_insert_lock(before, new_fl);
997 /* The new lock breaks the old one in two pieces,
998 * so we have to use the second new lock.
1002 locks_copy_lock(left, right);
1003 locks_insert_lock(before, left);
1005 right->fl_start = request->fl_end + 1;
1006 locks_wake_up_blocks(right);
1009 left->fl_end = request->fl_start - 1;
1010 locks_wake_up_blocks(left);
1015 * Free any unused locks.
1018 locks_free_lock(new_fl);
1020 locks_free_lock(new_fl2);
1025 * posix_lock_file - Apply a POSIX-style lock to a file
1026 * @filp: The file to apply the lock to
1027 * @fl: The lock to be applied
1028 * @conflock: Place to return a copy of the conflicting lock, if found.
1030 * Add a POSIX style lock to a file.
1031 * We merge adjacent & overlapping locks whenever possible.
1032 * POSIX locks are sorted by owner task, then by starting address
1034 * Note that if called with an FL_EXISTS argument, the caller may determine
1035 * whether or not a lock was successfully freed by testing the return
1036 * value for -ENOENT.
1038 int posix_lock_file(struct file *filp, struct file_lock *fl,
1039 struct file_lock *conflock)
1041 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1043 EXPORT_SYMBOL(posix_lock_file);
1046 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1047 * @filp: The file to apply the lock to
1048 * @fl: The lock to be applied
1050 * Add a POSIX style lock to a file.
1051 * We merge adjacent & overlapping locks whenever possible.
1052 * POSIX locks are sorted by owner task, then by starting address
1054 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1059 error = posix_lock_file(filp, fl, NULL);
1060 if (error != FILE_LOCK_DEFERRED)
1062 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1066 locks_delete_block(fl);
1071 EXPORT_SYMBOL(posix_lock_file_wait);
1074 * locks_mandatory_locked - Check for an active lock
1075 * @inode: the file to check
1077 * Searches the inode's list of locks to find any POSIX locks which conflict.
1078 * This function is called from locks_verify_locked() only.
1080 int locks_mandatory_locked(struct inode *inode)
1082 fl_owner_t owner = current->files;
1083 struct file_lock *fl;
1086 * Search the lock list for this inode for any POSIX locks.
1089 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1092 if (fl->fl_owner != owner)
1096 return fl ? -EAGAIN : 0;
1100 * locks_mandatory_area - Check for a conflicting lock
1101 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1103 * @inode: the file to check
1104 * @filp: how the file was opened (if it was)
1105 * @offset: start of area to check
1106 * @count: length of area to check
1108 * Searches the inode's list of locks to find any POSIX locks which conflict.
1109 * This function is called from rw_verify_area() and
1110 * locks_verify_truncate().
1112 int locks_mandatory_area(int read_write, struct inode *inode,
1113 struct file *filp, loff_t offset,
1116 struct file_lock fl;
1119 locks_init_lock(&fl);
1120 fl.fl_owner = current->files;
1121 fl.fl_pid = current->tgid;
1123 fl.fl_flags = FL_POSIX | FL_ACCESS;
1124 if (filp && !(filp->f_flags & O_NONBLOCK))
1125 fl.fl_flags |= FL_SLEEP;
1126 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1127 fl.fl_start = offset;
1128 fl.fl_end = offset + count - 1;
1131 error = __posix_lock_file(inode, &fl, NULL);
1132 if (error != FILE_LOCK_DEFERRED)
1134 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1137 * If we've been sleeping someone might have
1138 * changed the permissions behind our back.
1140 if (__mandatory_lock(inode))
1144 locks_delete_block(&fl);
1151 EXPORT_SYMBOL(locks_mandatory_area);
1153 /* We already had a lease on this file; just change its type */
1154 int lease_modify(struct file_lock **before, int arg)
1156 struct file_lock *fl = *before;
1157 int error = assign_type(fl, arg);
1161 locks_wake_up_blocks(fl);
1163 locks_delete_lock(before);
1167 EXPORT_SYMBOL(lease_modify);
1169 static void time_out_leases(struct inode *inode)
1171 struct file_lock **before;
1172 struct file_lock *fl;
1174 before = &inode->i_flock;
1175 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1176 if ((fl->fl_break_time == 0)
1177 || time_before(jiffies, fl->fl_break_time)) {
1178 before = &fl->fl_next;
1181 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1182 if (fl == *before) /* lease_modify may have freed fl */
1183 before = &fl->fl_next;
1188 * __break_lease - revoke all outstanding leases on file
1189 * @inode: the inode of the file to return
1190 * @mode: the open mode (read or write)
1192 * break_lease (inlined for speed) has checked there already is at least
1193 * some kind of lock (maybe a lease) on this file. Leases are broken on
1194 * a call to open() or truncate(). This function can sleep unless you
1195 * specified %O_NONBLOCK to your open().
1197 int __break_lease(struct inode *inode, unsigned int mode)
1199 int error = 0, future;
1200 struct file_lock *new_fl, *flock;
1201 struct file_lock *fl;
1202 unsigned long break_time;
1203 int i_have_this_lease = 0;
1204 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1206 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1210 time_out_leases(inode);
1212 flock = inode->i_flock;
1213 if ((flock == NULL) || !IS_LEASE(flock))
1216 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1217 if (fl->fl_owner == current->files)
1218 i_have_this_lease = 1;
1221 /* If we want write access, we have to revoke any lease. */
1222 future = F_UNLCK | F_INPROGRESS;
1223 } else if (flock->fl_type & F_INPROGRESS) {
1224 /* If the lease is already being broken, we just leave it */
1225 future = flock->fl_type;
1226 } else if (flock->fl_type & F_WRLCK) {
1227 /* Downgrade the exclusive lease to a read-only lease. */
1228 future = F_RDLCK | F_INPROGRESS;
1230 /* the existing lease was read-only, so we can read too. */
1234 if (IS_ERR(new_fl) && !i_have_this_lease
1235 && ((mode & O_NONBLOCK) == 0)) {
1236 error = PTR_ERR(new_fl);
1241 if (lease_break_time > 0) {
1242 break_time = jiffies + lease_break_time * HZ;
1243 if (break_time == 0)
1244 break_time++; /* so that 0 means no break time */
1247 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1248 if (fl->fl_type != future) {
1249 fl->fl_type = future;
1250 fl->fl_break_time = break_time;
1251 /* lease must have lmops break callback */
1252 fl->fl_lmops->fl_break(fl);
1256 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1257 error = -EWOULDBLOCK;
1262 break_time = flock->fl_break_time;
1263 if (break_time != 0) {
1264 break_time -= jiffies;
1265 if (break_time == 0)
1268 locks_insert_block(flock, new_fl);
1270 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1271 !new_fl->fl_next, break_time);
1273 __locks_delete_block(new_fl);
1276 time_out_leases(inode);
1277 /* Wait for the next lease that has not been broken yet */
1278 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1279 flock = flock->fl_next) {
1280 if (flock->fl_type & F_INPROGRESS)
1288 if (!IS_ERR(new_fl))
1289 locks_free_lock(new_fl);
1293 EXPORT_SYMBOL(__break_lease);
1296 * lease_get_mtime - get the last modified time of an inode
1298 * @time: pointer to a timespec which will contain the last modified time
1300 * This is to force NFS clients to flush their caches for files with
1301 * exclusive leases. The justification is that if someone has an
1302 * exclusive lease, then they could be modifying it.
1304 void lease_get_mtime(struct inode *inode, struct timespec *time)
1306 struct file_lock *flock = inode->i_flock;
1307 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1308 *time = current_fs_time(inode->i_sb);
1310 *time = inode->i_mtime;
1313 EXPORT_SYMBOL(lease_get_mtime);
1316 * fcntl_getlease - Enquire what lease is currently active
1319 * The value returned by this function will be one of
1320 * (if no lease break is pending):
1322 * %F_RDLCK to indicate a shared lease is held.
1324 * %F_WRLCK to indicate an exclusive lease is held.
1326 * %F_UNLCK to indicate no lease is held.
1328 * (if a lease break is pending):
1330 * %F_RDLCK to indicate an exclusive lease needs to be
1331 * changed to a shared lease (or removed).
1333 * %F_UNLCK to indicate the lease needs to be removed.
1335 * XXX: sfr & willy disagree over whether F_INPROGRESS
1336 * should be returned to userspace.
1338 int fcntl_getlease(struct file *filp)
1340 struct file_lock *fl;
1344 time_out_leases(filp->f_path.dentry->d_inode);
1345 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1347 if (fl->fl_file == filp) {
1348 type = fl->fl_type & ~F_INPROGRESS;
1357 * generic_setlease - sets a lease on an open file
1358 * @filp: file pointer
1359 * @arg: type of lease to obtain
1360 * @flp: input - file_lock to use, output - file_lock inserted
1362 * The (input) flp->fl_lmops->fl_break function is required
1365 * Called with file_lock_lock held.
1367 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1369 struct file_lock *fl, **before, **my_before = NULL, *lease;
1370 struct dentry *dentry = filp->f_path.dentry;
1371 struct inode *inode = dentry->d_inode;
1372 int error, rdlease_count = 0, wrlease_count = 0;
1377 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
1380 if (!S_ISREG(inode->i_mode))
1382 error = security_file_lock(filp, arg);
1386 time_out_leases(inode);
1388 BUG_ON(!(*flp)->fl_lmops->fl_break);
1390 if (arg != F_UNLCK) {
1392 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1394 if ((arg == F_WRLCK)
1395 && ((atomic_read(&dentry->d_count) > 1)
1396 || (atomic_read(&inode->i_count) > 1)))
1401 * At this point, we know that if there is an exclusive
1402 * lease on this file, then we hold it on this filp
1403 * (otherwise our open of this file would have blocked).
1404 * And if we are trying to acquire an exclusive lease,
1405 * then the file is not open by anyone (including us)
1406 * except for this filp.
1408 for (before = &inode->i_flock;
1409 ((fl = *before) != NULL) && IS_LEASE(fl);
1410 before = &fl->fl_next) {
1411 if (lease->fl_lmops->fl_mylease(fl, lease))
1413 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1415 * Someone is in the process of opening this
1416 * file for writing so we may not take an
1417 * exclusive lease on it.
1425 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1426 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1429 if (my_before != NULL) {
1431 error = lease->fl_lmops->fl_change(my_before, arg);
1442 locks_insert_lock(before, lease);
1447 locks_free_lock(lease);
1450 EXPORT_SYMBOL(generic_setlease);
1452 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1454 if (filp->f_op && filp->f_op->setlease)
1455 return filp->f_op->setlease(filp, arg, lease);
1457 return generic_setlease(filp, arg, lease);
1461 * vfs_setlease - sets a lease on an open file
1462 * @filp: file pointer
1463 * @arg: type of lease to obtain
1464 * @lease: file_lock to use
1466 * Call this to establish a lease on the file.
1467 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1468 * break_lease will oops!
1470 * This will call the filesystem's setlease file method, if
1471 * defined. Note that there is no getlease method; instead, the
1472 * filesystem setlease method should call back to setlease() to
1473 * add a lease to the inode's lease list, where fcntl_getlease() can
1474 * find it. Since fcntl_getlease() only reports whether the current
1475 * task holds a lease, a cluster filesystem need only do this for
1476 * leases held by processes on this node.
1478 * There is also no break_lease method; filesystems that
1479 * handle their own leases should break leases themselves from the
1480 * filesystem's open, create, and (on truncate) setattr methods.
1482 * Warning: the only current setlease methods exist only to disable
1483 * leases in certain cases. More vfs changes may be required to
1484 * allow a full filesystem lease implementation.
1487 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1492 error = __vfs_setlease(filp, arg, lease);
1497 EXPORT_SYMBOL_GPL(vfs_setlease);
1499 static int do_fcntl_delete_lease(struct file *filp)
1501 struct file_lock fl, *flp = &fl;
1503 lease_init(filp, F_UNLCK, flp);
1505 return vfs_setlease(filp, F_UNLCK, &flp);
1508 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1510 struct file_lock *fl;
1511 struct fasync_struct *new;
1512 struct inode *inode = filp->f_path.dentry->d_inode;
1515 fl = lease_alloc(filp, arg);
1519 new = fasync_alloc();
1521 locks_free_lock(fl);
1525 error = __vfs_setlease(filp, arg, &fl);
1530 * fasync_insert_entry() returns the old entry if any.
1531 * If there was no old entry, then it used 'new' and
1532 * inserted it into the fasync list. Clear new so that
1533 * we don't release it here.
1535 if (!fasync_insert_entry(fd, filp, &fl->fl_fasync, new))
1539 /* remove lease just inserted by setlease */
1540 fl->fl_type = F_UNLCK | F_INPROGRESS;
1541 fl->fl_break_time = jiffies - 10;
1542 time_out_leases(inode);
1546 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1555 * fcntl_setlease - sets a lease on an open file
1556 * @fd: open file descriptor
1557 * @filp: file pointer
1558 * @arg: type of lease to obtain
1560 * Call this fcntl to establish a lease on the file.
1561 * Note that you also need to call %F_SETSIG to
1562 * receive a signal when the lease is broken.
1564 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1567 return do_fcntl_delete_lease(filp);
1568 return do_fcntl_add_lease(fd, filp, arg);
1572 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1573 * @filp: The file to apply the lock to
1574 * @fl: The lock to be applied
1576 * Add a FLOCK style lock to a file.
1578 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1583 error = flock_lock_file(filp, fl);
1584 if (error != FILE_LOCK_DEFERRED)
1586 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1590 locks_delete_block(fl);
1596 EXPORT_SYMBOL(flock_lock_file_wait);
1599 * sys_flock: - flock() system call.
1600 * @fd: the file descriptor to lock.
1601 * @cmd: the type of lock to apply.
1603 * Apply a %FL_FLOCK style lock to an open file descriptor.
1604 * The @cmd can be one of
1606 * %LOCK_SH -- a shared lock.
1608 * %LOCK_EX -- an exclusive lock.
1610 * %LOCK_UN -- remove an existing lock.
1612 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1614 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1615 * processes read and write access respectively.
1617 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1620 struct file_lock *lock;
1621 int can_sleep, unlock;
1629 can_sleep = !(cmd & LOCK_NB);
1631 unlock = (cmd == LOCK_UN);
1633 if (!unlock && !(cmd & LOCK_MAND) &&
1634 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1637 error = flock_make_lock(filp, &lock, cmd);
1641 lock->fl_flags |= FL_SLEEP;
1643 error = security_file_lock(filp, lock->fl_type);
1647 if (filp->f_op && filp->f_op->flock)
1648 error = filp->f_op->flock(filp,
1649 (can_sleep) ? F_SETLKW : F_SETLK,
1652 error = flock_lock_file_wait(filp, lock);
1655 locks_free_lock(lock);
1664 * vfs_test_lock - test file byte range lock
1665 * @filp: The file to test lock for
1666 * @fl: The lock to test; also used to hold result
1668 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1669 * setting conf->fl_type to something other than F_UNLCK.
1671 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1673 if (filp->f_op && filp->f_op->lock)
1674 return filp->f_op->lock(filp, F_GETLK, fl);
1675 posix_test_lock(filp, fl);
1678 EXPORT_SYMBOL_GPL(vfs_test_lock);
1680 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1682 flock->l_pid = fl->fl_pid;
1683 #if BITS_PER_LONG == 32
1685 * Make sure we can represent the posix lock via
1686 * legacy 32bit flock.
1688 if (fl->fl_start > OFFT_OFFSET_MAX)
1690 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1693 flock->l_start = fl->fl_start;
1694 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1695 fl->fl_end - fl->fl_start + 1;
1696 flock->l_whence = 0;
1697 flock->l_type = fl->fl_type;
1701 #if BITS_PER_LONG == 32
1702 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1704 flock->l_pid = fl->fl_pid;
1705 flock->l_start = fl->fl_start;
1706 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1707 fl->fl_end - fl->fl_start + 1;
1708 flock->l_whence = 0;
1709 flock->l_type = fl->fl_type;
1713 /* Report the first existing lock that would conflict with l.
1714 * This implements the F_GETLK command of fcntl().
1716 int fcntl_getlk(struct file *filp, struct flock __user *l)
1718 struct file_lock file_lock;
1723 if (copy_from_user(&flock, l, sizeof(flock)))
1726 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1729 error = flock_to_posix_lock(filp, &file_lock, &flock);
1733 error = vfs_test_lock(filp, &file_lock);
1737 flock.l_type = file_lock.fl_type;
1738 if (file_lock.fl_type != F_UNLCK) {
1739 error = posix_lock_to_flock(&flock, &file_lock);
1744 if (!copy_to_user(l, &flock, sizeof(flock)))
1751 * vfs_lock_file - file byte range lock
1752 * @filp: The file to apply the lock to
1753 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1754 * @fl: The lock to be applied
1755 * @conf: Place to return a copy of the conflicting lock, if found.
1757 * A caller that doesn't care about the conflicting lock may pass NULL
1758 * as the final argument.
1760 * If the filesystem defines a private ->lock() method, then @conf will
1761 * be left unchanged; so a caller that cares should initialize it to
1762 * some acceptable default.
1764 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1765 * locks, the ->lock() interface may return asynchronously, before the lock has
1766 * been granted or denied by the underlying filesystem, if (and only if)
1767 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1768 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1769 * the request is for a blocking lock. When ->lock() does return asynchronously,
1770 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
1771 * request completes.
1772 * If the request is for non-blocking lock the file system should return
1773 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1774 * with the result. If the request timed out the callback routine will return a
1775 * nonzero return code and the file system should release the lock. The file
1776 * system is also responsible to keep a corresponding posix lock when it
1777 * grants a lock so the VFS can find out which locks are locally held and do
1778 * the correct lock cleanup when required.
1779 * The underlying filesystem must not drop the kernel lock or call
1780 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1783 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1785 if (filp->f_op && filp->f_op->lock)
1786 return filp->f_op->lock(filp, cmd, fl);
1788 return posix_lock_file(filp, fl, conf);
1790 EXPORT_SYMBOL_GPL(vfs_lock_file);
1792 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1793 struct file_lock *fl)
1797 error = security_file_lock(filp, fl->fl_type);
1802 error = vfs_lock_file(filp, cmd, fl, NULL);
1803 if (error != FILE_LOCK_DEFERRED)
1805 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1809 locks_delete_block(fl);
1816 /* Apply the lock described by l to an open file descriptor.
1817 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1819 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1820 struct flock __user *l)
1822 struct file_lock *file_lock = locks_alloc_lock();
1824 struct inode *inode;
1828 if (file_lock == NULL)
1832 * This might block, so we do it before checking the inode.
1835 if (copy_from_user(&flock, l, sizeof(flock)))
1838 inode = filp->f_path.dentry->d_inode;
1840 /* Don't allow mandatory locks on files that may be memory mapped
1843 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1849 error = flock_to_posix_lock(filp, file_lock, &flock);
1852 if (cmd == F_SETLKW) {
1853 file_lock->fl_flags |= FL_SLEEP;
1857 switch (flock.l_type) {
1859 if (!(filp->f_mode & FMODE_READ))
1863 if (!(filp->f_mode & FMODE_WRITE))
1873 error = do_lock_file_wait(filp, cmd, file_lock);
1876 * Attempt to detect a close/fcntl race and recover by
1877 * releasing the lock that was just acquired.
1880 * we need that spin_lock here - it prevents reordering between
1881 * update of inode->i_flock and check for it done in close().
1882 * rcu_read_lock() wouldn't do.
1884 spin_lock(¤t->files->file_lock);
1886 spin_unlock(¤t->files->file_lock);
1887 if (!error && f != filp && flock.l_type != F_UNLCK) {
1888 flock.l_type = F_UNLCK;
1893 locks_free_lock(file_lock);
1897 #if BITS_PER_LONG == 32
1898 /* Report the first existing lock that would conflict with l.
1899 * This implements the F_GETLK command of fcntl().
1901 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1903 struct file_lock file_lock;
1904 struct flock64 flock;
1908 if (copy_from_user(&flock, l, sizeof(flock)))
1911 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1914 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1918 error = vfs_test_lock(filp, &file_lock);
1922 flock.l_type = file_lock.fl_type;
1923 if (file_lock.fl_type != F_UNLCK)
1924 posix_lock_to_flock64(&flock, &file_lock);
1927 if (!copy_to_user(l, &flock, sizeof(flock)))
1934 /* Apply the lock described by l to an open file descriptor.
1935 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1937 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1938 struct flock64 __user *l)
1940 struct file_lock *file_lock = locks_alloc_lock();
1941 struct flock64 flock;
1942 struct inode *inode;
1946 if (file_lock == NULL)
1950 * This might block, so we do it before checking the inode.
1953 if (copy_from_user(&flock, l, sizeof(flock)))
1956 inode = filp->f_path.dentry->d_inode;
1958 /* Don't allow mandatory locks on files that may be memory mapped
1961 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1967 error = flock64_to_posix_lock(filp, file_lock, &flock);
1970 if (cmd == F_SETLKW64) {
1971 file_lock->fl_flags |= FL_SLEEP;
1975 switch (flock.l_type) {
1977 if (!(filp->f_mode & FMODE_READ))
1981 if (!(filp->f_mode & FMODE_WRITE))
1991 error = do_lock_file_wait(filp, cmd, file_lock);
1994 * Attempt to detect a close/fcntl race and recover by
1995 * releasing the lock that was just acquired.
1997 spin_lock(¤t->files->file_lock);
1999 spin_unlock(¤t->files->file_lock);
2000 if (!error && f != filp && flock.l_type != F_UNLCK) {
2001 flock.l_type = F_UNLCK;
2006 locks_free_lock(file_lock);
2009 #endif /* BITS_PER_LONG == 32 */
2012 * This function is called when the file is being removed
2013 * from the task's fd array. POSIX locks belonging to this task
2014 * are deleted at this time.
2016 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2018 struct file_lock lock;
2021 * If there are no locks held on this file, we don't need to call
2022 * posix_lock_file(). Another process could be setting a lock on this
2023 * file at the same time, but we wouldn't remove that lock anyway.
2025 if (!filp->f_path.dentry->d_inode->i_flock)
2028 lock.fl_type = F_UNLCK;
2029 lock.fl_flags = FL_POSIX | FL_CLOSE;
2031 lock.fl_end = OFFSET_MAX;
2032 lock.fl_owner = owner;
2033 lock.fl_pid = current->tgid;
2034 lock.fl_file = filp;
2036 lock.fl_lmops = NULL;
2038 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2040 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2041 lock.fl_ops->fl_release_private(&lock);
2044 EXPORT_SYMBOL(locks_remove_posix);
2047 * This function is called on the last close of an open file.
2049 void locks_remove_flock(struct file *filp)
2051 struct inode * inode = filp->f_path.dentry->d_inode;
2052 struct file_lock *fl;
2053 struct file_lock **before;
2055 if (!inode->i_flock)
2058 if (filp->f_op && filp->f_op->flock) {
2059 struct file_lock fl = {
2060 .fl_pid = current->tgid,
2062 .fl_flags = FL_FLOCK,
2064 .fl_end = OFFSET_MAX,
2066 filp->f_op->flock(filp, F_SETLKW, &fl);
2067 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2068 fl.fl_ops->fl_release_private(&fl);
2072 before = &inode->i_flock;
2074 while ((fl = *before) != NULL) {
2075 if (fl->fl_file == filp) {
2077 locks_delete_lock(before);
2081 lease_modify(before, F_UNLCK);
2087 before = &fl->fl_next;
2093 * posix_unblock_lock - stop waiting for a file lock
2094 * @filp: how the file was opened
2095 * @waiter: the lock which was waiting
2097 * lockd needs to block waiting for locks.
2100 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2105 if (waiter->fl_next)
2106 __locks_delete_block(waiter);
2113 EXPORT_SYMBOL(posix_unblock_lock);
2116 * vfs_cancel_lock - file byte range unblock lock
2117 * @filp: The file to apply the unblock to
2118 * @fl: The lock to be unblocked
2120 * Used by lock managers to cancel blocked requests
2122 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2124 if (filp->f_op && filp->f_op->lock)
2125 return filp->f_op->lock(filp, F_CANCELLK, fl);
2129 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2131 #ifdef CONFIG_PROC_FS
2132 #include <linux/proc_fs.h>
2133 #include <linux/seq_file.h>
2135 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2136 loff_t id, char *pfx)
2138 struct inode *inode = NULL;
2139 unsigned int fl_pid;
2142 fl_pid = pid_vnr(fl->fl_nspid);
2144 fl_pid = fl->fl_pid;
2146 if (fl->fl_file != NULL)
2147 inode = fl->fl_file->f_path.dentry->d_inode;
2149 seq_printf(f, "%lld:%s ", id, pfx);
2151 seq_printf(f, "%6s %s ",
2152 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2153 (inode == NULL) ? "*NOINODE*" :
2154 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2155 } else if (IS_FLOCK(fl)) {
2156 if (fl->fl_type & LOCK_MAND) {
2157 seq_printf(f, "FLOCK MSNFS ");
2159 seq_printf(f, "FLOCK ADVISORY ");
2161 } else if (IS_LEASE(fl)) {
2162 seq_printf(f, "LEASE ");
2163 if (fl->fl_type & F_INPROGRESS)
2164 seq_printf(f, "BREAKING ");
2165 else if (fl->fl_file)
2166 seq_printf(f, "ACTIVE ");
2168 seq_printf(f, "BREAKER ");
2170 seq_printf(f, "UNKNOWN UNKNOWN ");
2172 if (fl->fl_type & LOCK_MAND) {
2173 seq_printf(f, "%s ",
2174 (fl->fl_type & LOCK_READ)
2175 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2176 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2178 seq_printf(f, "%s ",
2179 (fl->fl_type & F_INPROGRESS)
2180 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2181 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2184 #ifdef WE_CAN_BREAK_LSLK_NOW
2185 seq_printf(f, "%d %s:%ld ", fl_pid,
2186 inode->i_sb->s_id, inode->i_ino);
2188 /* userspace relies on this representation of dev_t ;-( */
2189 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2190 MAJOR(inode->i_sb->s_dev),
2191 MINOR(inode->i_sb->s_dev), inode->i_ino);
2194 seq_printf(f, "%d <none>:0 ", fl_pid);
2197 if (fl->fl_end == OFFSET_MAX)
2198 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2200 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2202 seq_printf(f, "0 EOF\n");
2206 static int locks_show(struct seq_file *f, void *v)
2208 struct file_lock *fl, *bfl;
2210 fl = list_entry(v, struct file_lock, fl_link);
2212 lock_get_status(f, fl, *((loff_t *)f->private), "");
2214 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2215 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2220 static void *locks_start(struct seq_file *f, loff_t *pos)
2222 loff_t *p = f->private;
2226 return seq_list_start(&file_lock_list, *pos);
2229 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2231 loff_t *p = f->private;
2233 return seq_list_next(v, &file_lock_list, pos);
2236 static void locks_stop(struct seq_file *f, void *v)
2241 static const struct seq_operations locks_seq_operations = {
2242 .start = locks_start,
2248 static int locks_open(struct inode *inode, struct file *filp)
2250 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2253 static const struct file_operations proc_locks_operations = {
2256 .llseek = seq_lseek,
2257 .release = seq_release_private,
2260 static int __init proc_locks_init(void)
2262 proc_create("locks", 0, NULL, &proc_locks_operations);
2265 module_init(proc_locks_init);
2269 * lock_may_read - checks that the region is free of locks
2270 * @inode: the inode that is being read
2271 * @start: the first byte to read
2272 * @len: the number of bytes to read
2274 * Emulates Windows locking requirements. Whole-file
2275 * mandatory locks (share modes) can prohibit a read and
2276 * byte-range POSIX locks can prohibit a read if they overlap.
2278 * N.B. this function is only ever called
2279 * from knfsd and ownership of locks is never checked.
2281 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2283 struct file_lock *fl;
2286 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2288 if (fl->fl_type == F_RDLCK)
2290 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2292 } else if (IS_FLOCK(fl)) {
2293 if (!(fl->fl_type & LOCK_MAND))
2295 if (fl->fl_type & LOCK_READ)
2306 EXPORT_SYMBOL(lock_may_read);
2309 * lock_may_write - checks that the region is free of locks
2310 * @inode: the inode that is being written
2311 * @start: the first byte to write
2312 * @len: the number of bytes to write
2314 * Emulates Windows locking requirements. Whole-file
2315 * mandatory locks (share modes) can prohibit a write and
2316 * byte-range POSIX locks can prohibit a write if they overlap.
2318 * N.B. this function is only ever called
2319 * from knfsd and ownership of locks is never checked.
2321 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2323 struct file_lock *fl;
2326 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2328 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2330 } else if (IS_FLOCK(fl)) {
2331 if (!(fl->fl_type & LOCK_MAND))
2333 if (fl->fl_type & LOCK_WRITE)
2344 EXPORT_SYMBOL(lock_may_write);
2346 static int __init filelock_init(void)
2348 filelock_cache = kmem_cache_create("file_lock_cache",
2349 sizeof(struct file_lock), 0, SLAB_PANIC,
2354 core_initcall(filelock_init);