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/filesystems/mandatory-locking.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/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
128 #include <linux/pid_namespace.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 static bool lease_breaking(struct file_lock *fl)
138 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
141 static int target_leasetype(struct file_lock *fl)
143 if (fl->fl_flags & FL_UNLOCK_PENDING)
145 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
150 int leases_enable = 1;
151 int lease_break_time = 45;
153 #define for_each_lock(inode, lockp) \
154 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
156 /* The global file_lock_list is only used for displaying /proc/locks. */
157 static LIST_HEAD(file_lock_list);
159 /* The blocked_list is used to find POSIX lock loops for deadlock detection. */
160 static LIST_HEAD(blocked_list);
162 /* Protects the two list heads above, plus the inode->i_flock list */
163 static DEFINE_SPINLOCK(file_lock_lock);
165 void lock_flocks(void)
167 spin_lock(&file_lock_lock);
169 EXPORT_SYMBOL_GPL(lock_flocks);
171 void unlock_flocks(void)
173 spin_unlock(&file_lock_lock);
175 EXPORT_SYMBOL_GPL(unlock_flocks);
177 static struct kmem_cache *filelock_cache __read_mostly;
179 static void locks_init_lock_heads(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);
186 /* Allocate an empty lock structure. */
187 struct file_lock *locks_alloc_lock(void)
189 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
192 locks_init_lock_heads(fl);
196 EXPORT_SYMBOL_GPL(locks_alloc_lock);
198 void locks_release_private(struct file_lock *fl)
201 if (fl->fl_ops->fl_release_private)
202 fl->fl_ops->fl_release_private(fl);
208 EXPORT_SYMBOL_GPL(locks_release_private);
210 /* Free a lock which is not in use. */
211 void locks_free_lock(struct file_lock *fl)
213 BUG_ON(waitqueue_active(&fl->fl_wait));
214 BUG_ON(!list_empty(&fl->fl_block));
215 BUG_ON(!list_empty(&fl->fl_link));
217 locks_release_private(fl);
218 kmem_cache_free(filelock_cache, fl);
220 EXPORT_SYMBOL(locks_free_lock);
222 void locks_init_lock(struct file_lock *fl)
224 memset(fl, 0, sizeof(struct file_lock));
225 locks_init_lock_heads(fl);
228 EXPORT_SYMBOL(locks_init_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 new->fl_lmops = fl->fl_lmops;
242 * Initialize a new lock from an existing file_lock structure.
244 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
246 new->fl_owner = fl->fl_owner;
247 new->fl_pid = fl->fl_pid;
249 new->fl_flags = fl->fl_flags;
250 new->fl_type = fl->fl_type;
251 new->fl_start = fl->fl_start;
252 new->fl_end = fl->fl_end;
254 new->fl_lmops = NULL;
256 EXPORT_SYMBOL(__locks_copy_lock);
258 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
260 locks_release_private(new);
262 __locks_copy_lock(new, fl);
263 new->fl_file = fl->fl_file;
264 new->fl_ops = fl->fl_ops;
265 new->fl_lmops = fl->fl_lmops;
267 locks_copy_private(new, fl);
270 EXPORT_SYMBOL(locks_copy_lock);
272 static inline int flock_translate_cmd(int cmd) {
274 return cmd & (LOCK_MAND | LOCK_RW);
286 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
287 static int flock_make_lock(struct file *filp, struct file_lock **lock,
290 struct file_lock *fl;
291 int type = flock_translate_cmd(cmd);
295 fl = locks_alloc_lock();
300 fl->fl_pid = current->tgid;
301 fl->fl_flags = FL_FLOCK;
303 fl->fl_end = OFFSET_MAX;
309 static int assign_type(struct file_lock *fl, long type)
323 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
326 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
331 switch (l->l_whence) {
339 start = i_size_read(file_inode(filp));
345 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
346 POSIX-2001 defines it. */
350 fl->fl_end = OFFSET_MAX;
352 end = start + l->l_len - 1;
354 } else if (l->l_len < 0) {
361 fl->fl_start = start; /* we record the absolute position */
362 if (fl->fl_end < fl->fl_start)
365 fl->fl_owner = current->files;
366 fl->fl_pid = current->tgid;
368 fl->fl_flags = FL_POSIX;
372 return assign_type(fl, l->l_type);
375 #if BITS_PER_LONG == 32
376 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
381 switch (l->l_whence) {
389 start = i_size_read(file_inode(filp));
398 fl->fl_end = OFFSET_MAX;
400 fl->fl_end = start + l->l_len - 1;
401 } else if (l->l_len < 0) {
402 fl->fl_end = start - 1;
407 fl->fl_start = start; /* we record the absolute position */
408 if (fl->fl_end < fl->fl_start)
411 fl->fl_owner = current->files;
412 fl->fl_pid = current->tgid;
414 fl->fl_flags = FL_POSIX;
418 return assign_type(fl, l->l_type);
422 /* default lease lock manager operations */
423 static void lease_break_callback(struct file_lock *fl)
425 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
428 static const struct lock_manager_operations lease_manager_ops = {
429 .lm_break = lease_break_callback,
430 .lm_change = lease_modify,
434 * Initialize a lease, use the default lock manager operations
436 static int lease_init(struct file *filp, long type, struct file_lock *fl)
438 if (assign_type(fl, type) != 0)
441 fl->fl_owner = current->files;
442 fl->fl_pid = current->tgid;
445 fl->fl_flags = FL_LEASE;
447 fl->fl_end = OFFSET_MAX;
449 fl->fl_lmops = &lease_manager_ops;
453 /* Allocate a file_lock initialised to this type of lease */
454 static struct file_lock *lease_alloc(struct file *filp, long type)
456 struct file_lock *fl = locks_alloc_lock();
460 return ERR_PTR(error);
462 error = lease_init(filp, type, fl);
465 return ERR_PTR(error);
470 /* Check if two locks overlap each other.
472 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
474 return ((fl1->fl_end >= fl2->fl_start) &&
475 (fl2->fl_end >= fl1->fl_start));
479 * Check whether two locks have the same owner.
481 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
483 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
484 return fl2->fl_lmops == fl1->fl_lmops &&
485 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
486 return fl1->fl_owner == fl2->fl_owner;
490 locks_insert_global_locks(struct file_lock *fl)
492 list_add_tail(&fl->fl_link, &file_lock_list);
496 locks_delete_global_locks(struct file_lock *fl)
498 list_del_init(&fl->fl_link);
502 locks_insert_global_blocked(struct file_lock *waiter)
504 list_add(&waiter->fl_link, &blocked_list);
508 locks_delete_global_blocked(struct file_lock *waiter)
510 list_del_init(&waiter->fl_link);
513 /* Remove waiter from blocker's block list.
514 * When blocker ends up pointing to itself then the list is empty.
516 static void __locks_delete_block(struct file_lock *waiter)
518 locks_delete_global_blocked(waiter);
519 list_del_init(&waiter->fl_block);
520 waiter->fl_next = NULL;
525 static void locks_delete_block(struct file_lock *waiter)
528 __locks_delete_block(waiter);
532 /* Insert waiter into blocker's block list.
533 * We use a circular list so that processes can be easily woken up in
534 * the order they blocked. The documentation doesn't require this but
535 * it seems like the reasonable thing to do.
537 static void locks_insert_block(struct file_lock *blocker,
538 struct file_lock *waiter)
540 BUG_ON(!list_empty(&waiter->fl_block));
541 waiter->fl_next = blocker;
542 list_add_tail(&waiter->fl_block, &blocker->fl_block);
543 if (IS_POSIX(blocker))
544 locks_insert_global_blocked(request);
548 * Wake up processes blocked waiting for blocker.
550 * Must be called with the file_lock_lock held!
552 static void locks_wake_up_blocks(struct file_lock *blocker)
554 while (!list_empty(&blocker->fl_block)) {
555 struct file_lock *waiter;
557 waiter = list_first_entry(&blocker->fl_block,
558 struct file_lock, fl_block);
559 __locks_delete_block(waiter);
560 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
561 waiter->fl_lmops->lm_notify(waiter);
563 wake_up(&waiter->fl_wait);
567 /* Insert file lock fl into an inode's lock list at the position indicated
568 * by pos. At the same time add the lock to the global file lock list.
570 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
572 fl->fl_nspid = get_pid(task_tgid(current));
574 /* insert into file's list */
578 locks_insert_global_locks(fl);
582 * Delete a lock and then free it.
583 * Wake up processes that are blocked waiting for this lock,
584 * notify the FS that the lock has been cleared and
585 * finally free the lock.
587 static void locks_delete_lock(struct file_lock **thisfl_p)
589 struct file_lock *fl = *thisfl_p;
591 locks_delete_global_locks(fl);
593 *thisfl_p = fl->fl_next;
597 put_pid(fl->fl_nspid);
601 locks_wake_up_blocks(fl);
605 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
606 * checks for shared/exclusive status of overlapping locks.
608 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
610 if (sys_fl->fl_type == F_WRLCK)
612 if (caller_fl->fl_type == F_WRLCK)
617 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
618 * checking before calling the locks_conflict().
620 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
622 /* POSIX locks owned by the same process do not conflict with
625 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
628 /* Check whether they overlap */
629 if (!locks_overlap(caller_fl, sys_fl))
632 return (locks_conflict(caller_fl, sys_fl));
635 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
636 * checking before calling the locks_conflict().
638 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
640 /* FLOCK locks referring to the same filp do not conflict with
643 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
645 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
648 return (locks_conflict(caller_fl, sys_fl));
652 posix_test_lock(struct file *filp, struct file_lock *fl)
654 struct file_lock *cfl;
657 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
660 if (posix_locks_conflict(fl, cfl))
664 __locks_copy_lock(fl, cfl);
666 fl->fl_pid = pid_vnr(cfl->fl_nspid);
668 fl->fl_type = F_UNLCK;
672 EXPORT_SYMBOL(posix_test_lock);
675 * Deadlock detection:
677 * We attempt to detect deadlocks that are due purely to posix file
680 * We assume that a task can be waiting for at most one lock at a time.
681 * So for any acquired lock, the process holding that lock may be
682 * waiting on at most one other lock. That lock in turns may be held by
683 * someone waiting for at most one other lock. Given a requested lock
684 * caller_fl which is about to wait for a conflicting lock block_fl, we
685 * follow this chain of waiters to ensure we are not about to create a
688 * Since we do this before we ever put a process to sleep on a lock, we
689 * are ensured that there is never a cycle; that is what guarantees that
690 * the while() loop in posix_locks_deadlock() eventually completes.
692 * Note: the above assumption may not be true when handling lock
693 * requests from a broken NFS client. It may also fail in the presence
694 * of tasks (such as posix threads) sharing the same open file table.
696 * To handle those cases, we just bail out after a few iterations.
699 #define MAX_DEADLK_ITERATIONS 10
701 /* Find a lock that the owner of the given block_fl is blocking on. */
702 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
704 struct file_lock *fl;
706 list_for_each_entry(fl, &blocked_list, fl_link) {
707 if (posix_same_owner(fl, block_fl))
713 static int posix_locks_deadlock(struct file_lock *caller_fl,
714 struct file_lock *block_fl)
718 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
719 if (i++ > MAX_DEADLK_ITERATIONS)
721 if (posix_same_owner(caller_fl, block_fl))
727 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
728 * after any leases, but before any posix locks.
730 * Note that if called with an FL_EXISTS argument, the caller may determine
731 * whether or not a lock was successfully freed by testing the return
734 static int flock_lock_file(struct file *filp, struct file_lock *request)
736 struct file_lock *new_fl = NULL;
737 struct file_lock **before;
738 struct inode * inode = file_inode(filp);
742 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
743 new_fl = locks_alloc_lock();
749 if (request->fl_flags & FL_ACCESS)
752 for_each_lock(inode, before) {
753 struct file_lock *fl = *before;
758 if (filp != fl->fl_file)
760 if (request->fl_type == fl->fl_type)
763 locks_delete_lock(before);
767 if (request->fl_type == F_UNLCK) {
768 if ((request->fl_flags & FL_EXISTS) && !found)
774 * If a higher-priority process was blocked on the old file lock,
775 * give it the opportunity to lock the file.
784 for_each_lock(inode, before) {
785 struct file_lock *fl = *before;
790 if (!flock_locks_conflict(request, fl))
793 if (!(request->fl_flags & FL_SLEEP))
795 error = FILE_LOCK_DEFERRED;
796 locks_insert_block(fl, request);
799 if (request->fl_flags & FL_ACCESS)
801 locks_copy_lock(new_fl, request);
802 locks_insert_lock(before, new_fl);
809 locks_free_lock(new_fl);
813 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
815 struct file_lock *fl;
816 struct file_lock *new_fl = NULL;
817 struct file_lock *new_fl2 = NULL;
818 struct file_lock *left = NULL;
819 struct file_lock *right = NULL;
820 struct file_lock **before;
825 * We may need two file_lock structures for this operation,
826 * so we get them in advance to avoid races.
828 * In some cases we can be sure, that no new locks will be needed
830 if (!(request->fl_flags & FL_ACCESS) &&
831 (request->fl_type != F_UNLCK ||
832 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
833 new_fl = locks_alloc_lock();
834 new_fl2 = locks_alloc_lock();
839 * New lock request. Walk all POSIX locks and look for conflicts. If
840 * there are any, either return error or put the request on the
841 * blocker's list of waiters and the global blocked_list.
843 if (request->fl_type != F_UNLCK) {
844 for_each_lock(inode, before) {
848 if (!posix_locks_conflict(request, fl))
851 __locks_copy_lock(conflock, fl);
853 if (!(request->fl_flags & FL_SLEEP))
856 if (posix_locks_deadlock(request, fl))
858 error = FILE_LOCK_DEFERRED;
859 locks_insert_block(fl, request);
864 /* If we're just looking for a conflict, we're done. */
866 if (request->fl_flags & FL_ACCESS)
870 * Find the first old lock with the same owner as the new lock.
873 before = &inode->i_flock;
875 /* First skip locks owned by other processes. */
876 while ((fl = *before) && (!IS_POSIX(fl) ||
877 !posix_same_owner(request, fl))) {
878 before = &fl->fl_next;
881 /* Process locks with this owner. */
882 while ((fl = *before) && posix_same_owner(request, fl)) {
883 /* Detect adjacent or overlapping regions (if same lock type)
885 if (request->fl_type == fl->fl_type) {
886 /* In all comparisons of start vs end, use
887 * "start - 1" rather than "end + 1". If end
888 * is OFFSET_MAX, end + 1 will become negative.
890 if (fl->fl_end < request->fl_start - 1)
892 /* If the next lock in the list has entirely bigger
893 * addresses than the new one, insert the lock here.
895 if (fl->fl_start - 1 > request->fl_end)
898 /* If we come here, the new and old lock are of the
899 * same type and adjacent or overlapping. Make one
900 * lock yielding from the lower start address of both
901 * locks to the higher end address.
903 if (fl->fl_start > request->fl_start)
904 fl->fl_start = request->fl_start;
906 request->fl_start = fl->fl_start;
907 if (fl->fl_end < request->fl_end)
908 fl->fl_end = request->fl_end;
910 request->fl_end = fl->fl_end;
912 locks_delete_lock(before);
919 /* Processing for different lock types is a bit
922 if (fl->fl_end < request->fl_start)
924 if (fl->fl_start > request->fl_end)
926 if (request->fl_type == F_UNLCK)
928 if (fl->fl_start < request->fl_start)
930 /* If the next lock in the list has a higher end
931 * address than the new one, insert the new one here.
933 if (fl->fl_end > request->fl_end) {
937 if (fl->fl_start >= request->fl_start) {
938 /* The new lock completely replaces an old
939 * one (This may happen several times).
942 locks_delete_lock(before);
945 /* Replace the old lock with the new one.
946 * Wake up anybody waiting for the old one,
947 * as the change in lock type might satisfy
950 locks_wake_up_blocks(fl);
951 fl->fl_start = request->fl_start;
952 fl->fl_end = request->fl_end;
953 fl->fl_type = request->fl_type;
954 locks_release_private(fl);
955 locks_copy_private(fl, request);
960 /* Go on to next lock.
963 before = &fl->fl_next;
967 * The above code only modifies existing locks in case of merging or
968 * replacing. If new lock(s) need to be inserted all modifications are
969 * done below this, so it's safe yet to bail out.
971 error = -ENOLCK; /* "no luck" */
972 if (right && left == right && !new_fl2)
977 if (request->fl_type == F_UNLCK) {
978 if (request->fl_flags & FL_EXISTS)
987 locks_copy_lock(new_fl, request);
988 locks_insert_lock(before, new_fl);
993 /* The new lock breaks the old one in two pieces,
994 * so we have to use the second new lock.
998 locks_copy_lock(left, right);
999 locks_insert_lock(before, left);
1001 right->fl_start = request->fl_end + 1;
1002 locks_wake_up_blocks(right);
1005 left->fl_end = request->fl_start - 1;
1006 locks_wake_up_blocks(left);
1011 * Free any unused locks.
1014 locks_free_lock(new_fl);
1016 locks_free_lock(new_fl2);
1021 * posix_lock_file - Apply a POSIX-style lock to a file
1022 * @filp: The file to apply the lock to
1023 * @fl: The lock to be applied
1024 * @conflock: Place to return a copy of the conflicting lock, if found.
1026 * Add a POSIX style lock to a file.
1027 * We merge adjacent & overlapping locks whenever possible.
1028 * POSIX locks are sorted by owner task, then by starting address
1030 * Note that if called with an FL_EXISTS argument, the caller may determine
1031 * whether or not a lock was successfully freed by testing the return
1032 * value for -ENOENT.
1034 int posix_lock_file(struct file *filp, struct file_lock *fl,
1035 struct file_lock *conflock)
1037 return __posix_lock_file(file_inode(filp), fl, conflock);
1039 EXPORT_SYMBOL(posix_lock_file);
1042 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1043 * @filp: The file to apply the lock to
1044 * @fl: The lock to be applied
1046 * Add a POSIX style lock to a file.
1047 * We merge adjacent & overlapping locks whenever possible.
1048 * POSIX locks are sorted by owner task, then by starting address
1050 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1055 error = posix_lock_file(filp, fl, NULL);
1056 if (error != FILE_LOCK_DEFERRED)
1058 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1062 locks_delete_block(fl);
1067 EXPORT_SYMBOL(posix_lock_file_wait);
1070 * locks_mandatory_locked - Check for an active lock
1071 * @inode: the file to check
1073 * Searches the inode's list of locks to find any POSIX locks which conflict.
1074 * This function is called from locks_verify_locked() only.
1076 int locks_mandatory_locked(struct inode *inode)
1078 fl_owner_t owner = current->files;
1079 struct file_lock *fl;
1082 * Search the lock list for this inode for any POSIX locks.
1085 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1088 if (fl->fl_owner != owner)
1092 return fl ? -EAGAIN : 0;
1096 * locks_mandatory_area - Check for a conflicting lock
1097 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1099 * @inode: the file to check
1100 * @filp: how the file was opened (if it was)
1101 * @offset: start of area to check
1102 * @count: length of area to check
1104 * Searches the inode's list of locks to find any POSIX locks which conflict.
1105 * This function is called from rw_verify_area() and
1106 * locks_verify_truncate().
1108 int locks_mandatory_area(int read_write, struct inode *inode,
1109 struct file *filp, loff_t offset,
1112 struct file_lock fl;
1115 locks_init_lock(&fl);
1116 fl.fl_owner = current->files;
1117 fl.fl_pid = current->tgid;
1119 fl.fl_flags = FL_POSIX | FL_ACCESS;
1120 if (filp && !(filp->f_flags & O_NONBLOCK))
1121 fl.fl_flags |= FL_SLEEP;
1122 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1123 fl.fl_start = offset;
1124 fl.fl_end = offset + count - 1;
1127 error = __posix_lock_file(inode, &fl, NULL);
1128 if (error != FILE_LOCK_DEFERRED)
1130 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1133 * If we've been sleeping someone might have
1134 * changed the permissions behind our back.
1136 if (__mandatory_lock(inode))
1140 locks_delete_block(&fl);
1147 EXPORT_SYMBOL(locks_mandatory_area);
1149 static void lease_clear_pending(struct file_lock *fl, int arg)
1153 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1156 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1160 /* We already had a lease on this file; just change its type */
1161 int lease_modify(struct file_lock **before, int arg)
1163 struct file_lock *fl = *before;
1164 int error = assign_type(fl, arg);
1168 lease_clear_pending(fl, arg);
1169 locks_wake_up_blocks(fl);
1170 if (arg == F_UNLCK) {
1171 struct file *filp = fl->fl_file;
1174 filp->f_owner.signum = 0;
1175 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1176 if (fl->fl_fasync != NULL) {
1177 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1178 fl->fl_fasync = NULL;
1180 locks_delete_lock(before);
1185 EXPORT_SYMBOL(lease_modify);
1187 static bool past_time(unsigned long then)
1190 /* 0 is a special value meaning "this never expires": */
1192 return time_after(jiffies, then);
1195 static void time_out_leases(struct inode *inode)
1197 struct file_lock **before;
1198 struct file_lock *fl;
1200 before = &inode->i_flock;
1201 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1202 if (past_time(fl->fl_downgrade_time))
1203 lease_modify(before, F_RDLCK);
1204 if (past_time(fl->fl_break_time))
1205 lease_modify(before, F_UNLCK);
1206 if (fl == *before) /* lease_modify may have freed fl */
1207 before = &fl->fl_next;
1212 * __break_lease - revoke all outstanding leases on file
1213 * @inode: the inode of the file to return
1214 * @mode: the open mode (read or write)
1216 * break_lease (inlined for speed) has checked there already is at least
1217 * some kind of lock (maybe a lease) on this file. Leases are broken on
1218 * a call to open() or truncate(). This function can sleep unless you
1219 * specified %O_NONBLOCK to your open().
1221 int __break_lease(struct inode *inode, unsigned int mode)
1224 struct file_lock *new_fl, *flock;
1225 struct file_lock *fl;
1226 unsigned long break_time;
1227 int i_have_this_lease = 0;
1228 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1230 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1232 return PTR_ERR(new_fl);
1236 time_out_leases(inode);
1238 flock = inode->i_flock;
1239 if ((flock == NULL) || !IS_LEASE(flock))
1242 if (!locks_conflict(flock, new_fl))
1245 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1246 if (fl->fl_owner == current->files)
1247 i_have_this_lease = 1;
1250 if (lease_break_time > 0) {
1251 break_time = jiffies + lease_break_time * HZ;
1252 if (break_time == 0)
1253 break_time++; /* so that 0 means no break time */
1256 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1258 if (fl->fl_flags & FL_UNLOCK_PENDING)
1260 fl->fl_flags |= FL_UNLOCK_PENDING;
1261 fl->fl_break_time = break_time;
1263 if (lease_breaking(flock))
1265 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1266 fl->fl_downgrade_time = break_time;
1268 fl->fl_lmops->lm_break(fl);
1271 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1272 error = -EWOULDBLOCK;
1277 break_time = flock->fl_break_time;
1278 if (break_time != 0) {
1279 break_time -= jiffies;
1280 if (break_time == 0)
1283 locks_insert_block(flock, new_fl);
1285 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1286 !new_fl->fl_next, break_time);
1288 __locks_delete_block(new_fl);
1291 time_out_leases(inode);
1293 * Wait for the next conflicting lease that has not been
1296 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1297 flock = flock->fl_next) {
1298 if (locks_conflict(new_fl, flock))
1306 locks_free_lock(new_fl);
1310 EXPORT_SYMBOL(__break_lease);
1313 * lease_get_mtime - get the last modified time of an inode
1315 * @time: pointer to a timespec which will contain the last modified time
1317 * This is to force NFS clients to flush their caches for files with
1318 * exclusive leases. The justification is that if someone has an
1319 * exclusive lease, then they could be modifying it.
1321 void lease_get_mtime(struct inode *inode, struct timespec *time)
1323 struct file_lock *flock = inode->i_flock;
1324 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1325 *time = current_fs_time(inode->i_sb);
1327 *time = inode->i_mtime;
1330 EXPORT_SYMBOL(lease_get_mtime);
1333 * fcntl_getlease - Enquire what lease is currently active
1336 * The value returned by this function will be one of
1337 * (if no lease break is pending):
1339 * %F_RDLCK to indicate a shared lease is held.
1341 * %F_WRLCK to indicate an exclusive lease is held.
1343 * %F_UNLCK to indicate no lease is held.
1345 * (if a lease break is pending):
1347 * %F_RDLCK to indicate an exclusive lease needs to be
1348 * changed to a shared lease (or removed).
1350 * %F_UNLCK to indicate the lease needs to be removed.
1352 * XXX: sfr & willy disagree over whether F_INPROGRESS
1353 * should be returned to userspace.
1355 int fcntl_getlease(struct file *filp)
1357 struct file_lock *fl;
1361 time_out_leases(file_inode(filp));
1362 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1364 if (fl->fl_file == filp) {
1365 type = target_leasetype(fl);
1373 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1375 struct file_lock *fl, **before, **my_before = NULL, *lease;
1376 struct dentry *dentry = filp->f_path.dentry;
1377 struct inode *inode = dentry->d_inode;
1383 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1385 if ((arg == F_WRLCK)
1386 && ((dentry->d_count > 1)
1387 || (atomic_read(&inode->i_count) > 1)))
1391 * At this point, we know that if there is an exclusive
1392 * lease on this file, then we hold it on this filp
1393 * (otherwise our open of this file would have blocked).
1394 * And if we are trying to acquire an exclusive lease,
1395 * then the file is not open by anyone (including us)
1396 * except for this filp.
1399 for (before = &inode->i_flock;
1400 ((fl = *before) != NULL) && IS_LEASE(fl);
1401 before = &fl->fl_next) {
1402 if (fl->fl_file == filp) {
1407 * No exclusive leases if someone else has a lease on
1413 * Modifying our existing lease is OK, but no getting a
1414 * new lease if someone else is opening for write:
1416 if (fl->fl_flags & FL_UNLOCK_PENDING)
1420 if (my_before != NULL) {
1421 error = lease->fl_lmops->lm_change(my_before, arg);
1431 locks_insert_lock(before, lease);
1438 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1440 struct file_lock *fl, **before;
1441 struct dentry *dentry = filp->f_path.dentry;
1442 struct inode *inode = dentry->d_inode;
1444 for (before = &inode->i_flock;
1445 ((fl = *before) != NULL) && IS_LEASE(fl);
1446 before = &fl->fl_next) {
1447 if (fl->fl_file != filp)
1449 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1455 * generic_setlease - sets a lease on an open file
1456 * @filp: file pointer
1457 * @arg: type of lease to obtain
1458 * @flp: input - file_lock to use, output - file_lock inserted
1460 * The (input) flp->fl_lmops->lm_break function is required
1463 * Called with file_lock_lock held.
1465 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1467 struct dentry *dentry = filp->f_path.dentry;
1468 struct inode *inode = dentry->d_inode;
1471 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1473 if (!S_ISREG(inode->i_mode))
1475 error = security_file_lock(filp, arg);
1479 time_out_leases(inode);
1481 BUG_ON(!(*flp)->fl_lmops->lm_break);
1485 return generic_delete_lease(filp, flp);
1488 return generic_add_lease(filp, arg, flp);
1493 EXPORT_SYMBOL(generic_setlease);
1495 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1497 if (filp->f_op && filp->f_op->setlease)
1498 return filp->f_op->setlease(filp, arg, lease);
1500 return generic_setlease(filp, arg, lease);
1504 * vfs_setlease - sets a lease on an open file
1505 * @filp: file pointer
1506 * @arg: type of lease to obtain
1507 * @lease: file_lock to use
1509 * Call this to establish a lease on the file.
1510 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1511 * break_lease will oops!
1513 * This will call the filesystem's setlease file method, if
1514 * defined. Note that there is no getlease method; instead, the
1515 * filesystem setlease method should call back to setlease() to
1516 * add a lease to the inode's lease list, where fcntl_getlease() can
1517 * find it. Since fcntl_getlease() only reports whether the current
1518 * task holds a lease, a cluster filesystem need only do this for
1519 * leases held by processes on this node.
1521 * There is also no break_lease method; filesystems that
1522 * handle their own leases should break leases themselves from the
1523 * filesystem's open, create, and (on truncate) setattr methods.
1525 * Warning: the only current setlease methods exist only to disable
1526 * leases in certain cases. More vfs changes may be required to
1527 * allow a full filesystem lease implementation.
1530 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1535 error = __vfs_setlease(filp, arg, lease);
1540 EXPORT_SYMBOL_GPL(vfs_setlease);
1542 static int do_fcntl_delete_lease(struct file *filp)
1544 struct file_lock fl, *flp = &fl;
1546 lease_init(filp, F_UNLCK, flp);
1548 return vfs_setlease(filp, F_UNLCK, &flp);
1551 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1553 struct file_lock *fl, *ret;
1554 struct fasync_struct *new;
1557 fl = lease_alloc(filp, arg);
1561 new = fasync_alloc();
1563 locks_free_lock(fl);
1568 error = __vfs_setlease(filp, arg, &ret);
1571 locks_free_lock(fl);
1572 goto out_free_fasync;
1575 locks_free_lock(fl);
1578 * fasync_insert_entry() returns the old entry if any.
1579 * If there was no old entry, then it used 'new' and
1580 * inserted it into the fasync list. Clear new so that
1581 * we don't release it here.
1583 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1586 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1596 * fcntl_setlease - sets a lease on an open file
1597 * @fd: open file descriptor
1598 * @filp: file pointer
1599 * @arg: type of lease to obtain
1601 * Call this fcntl to establish a lease on the file.
1602 * Note that you also need to call %F_SETSIG to
1603 * receive a signal when the lease is broken.
1605 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1608 return do_fcntl_delete_lease(filp);
1609 return do_fcntl_add_lease(fd, filp, arg);
1613 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1614 * @filp: The file to apply the lock to
1615 * @fl: The lock to be applied
1617 * Add a FLOCK style lock to a file.
1619 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1624 error = flock_lock_file(filp, fl);
1625 if (error != FILE_LOCK_DEFERRED)
1627 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1631 locks_delete_block(fl);
1637 EXPORT_SYMBOL(flock_lock_file_wait);
1640 * sys_flock: - flock() system call.
1641 * @fd: the file descriptor to lock.
1642 * @cmd: the type of lock to apply.
1644 * Apply a %FL_FLOCK style lock to an open file descriptor.
1645 * The @cmd can be one of
1647 * %LOCK_SH -- a shared lock.
1649 * %LOCK_EX -- an exclusive lock.
1651 * %LOCK_UN -- remove an existing lock.
1653 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1655 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1656 * processes read and write access respectively.
1658 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1660 struct fd f = fdget(fd);
1661 struct file_lock *lock;
1662 int can_sleep, unlock;
1669 can_sleep = !(cmd & LOCK_NB);
1671 unlock = (cmd == LOCK_UN);
1673 if (!unlock && !(cmd & LOCK_MAND) &&
1674 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1677 error = flock_make_lock(f.file, &lock, cmd);
1681 lock->fl_flags |= FL_SLEEP;
1683 error = security_file_lock(f.file, lock->fl_type);
1687 if (f.file->f_op && f.file->f_op->flock)
1688 error = f.file->f_op->flock(f.file,
1689 (can_sleep) ? F_SETLKW : F_SETLK,
1692 error = flock_lock_file_wait(f.file, lock);
1695 locks_free_lock(lock);
1704 * vfs_test_lock - test file byte range lock
1705 * @filp: The file to test lock for
1706 * @fl: The lock to test; also used to hold result
1708 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1709 * setting conf->fl_type to something other than F_UNLCK.
1711 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1713 if (filp->f_op && filp->f_op->lock)
1714 return filp->f_op->lock(filp, F_GETLK, fl);
1715 posix_test_lock(filp, fl);
1718 EXPORT_SYMBOL_GPL(vfs_test_lock);
1720 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1722 flock->l_pid = fl->fl_pid;
1723 #if BITS_PER_LONG == 32
1725 * Make sure we can represent the posix lock via
1726 * legacy 32bit flock.
1728 if (fl->fl_start > OFFT_OFFSET_MAX)
1730 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1733 flock->l_start = fl->fl_start;
1734 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1735 fl->fl_end - fl->fl_start + 1;
1736 flock->l_whence = 0;
1737 flock->l_type = fl->fl_type;
1741 #if BITS_PER_LONG == 32
1742 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1744 flock->l_pid = fl->fl_pid;
1745 flock->l_start = fl->fl_start;
1746 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1747 fl->fl_end - fl->fl_start + 1;
1748 flock->l_whence = 0;
1749 flock->l_type = fl->fl_type;
1753 /* Report the first existing lock that would conflict with l.
1754 * This implements the F_GETLK command of fcntl().
1756 int fcntl_getlk(struct file *filp, struct flock __user *l)
1758 struct file_lock file_lock;
1763 if (copy_from_user(&flock, l, sizeof(flock)))
1766 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1769 error = flock_to_posix_lock(filp, &file_lock, &flock);
1773 error = vfs_test_lock(filp, &file_lock);
1777 flock.l_type = file_lock.fl_type;
1778 if (file_lock.fl_type != F_UNLCK) {
1779 error = posix_lock_to_flock(&flock, &file_lock);
1784 if (!copy_to_user(l, &flock, sizeof(flock)))
1791 * vfs_lock_file - file byte range lock
1792 * @filp: The file to apply the lock to
1793 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1794 * @fl: The lock to be applied
1795 * @conf: Place to return a copy of the conflicting lock, if found.
1797 * A caller that doesn't care about the conflicting lock may pass NULL
1798 * as the final argument.
1800 * If the filesystem defines a private ->lock() method, then @conf will
1801 * be left unchanged; so a caller that cares should initialize it to
1802 * some acceptable default.
1804 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1805 * locks, the ->lock() interface may return asynchronously, before the lock has
1806 * been granted or denied by the underlying filesystem, if (and only if)
1807 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1808 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1809 * the request is for a blocking lock. When ->lock() does return asynchronously,
1810 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1811 * request completes.
1812 * If the request is for non-blocking lock the file system should return
1813 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1814 * with the result. If the request timed out the callback routine will return a
1815 * nonzero return code and the file system should release the lock. The file
1816 * system is also responsible to keep a corresponding posix lock when it
1817 * grants a lock so the VFS can find out which locks are locally held and do
1818 * the correct lock cleanup when required.
1819 * The underlying filesystem must not drop the kernel lock or call
1820 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1823 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1825 if (filp->f_op && filp->f_op->lock)
1826 return filp->f_op->lock(filp, cmd, fl);
1828 return posix_lock_file(filp, fl, conf);
1830 EXPORT_SYMBOL_GPL(vfs_lock_file);
1832 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1833 struct file_lock *fl)
1837 error = security_file_lock(filp, fl->fl_type);
1842 error = vfs_lock_file(filp, cmd, fl, NULL);
1843 if (error != FILE_LOCK_DEFERRED)
1845 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1849 locks_delete_block(fl);
1856 /* Apply the lock described by l to an open file descriptor.
1857 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1859 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1860 struct flock __user *l)
1862 struct file_lock *file_lock = locks_alloc_lock();
1864 struct inode *inode;
1868 if (file_lock == NULL)
1872 * This might block, so we do it before checking the inode.
1875 if (copy_from_user(&flock, l, sizeof(flock)))
1878 inode = file_inode(filp);
1880 /* Don't allow mandatory locks on files that may be memory mapped
1883 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1889 error = flock_to_posix_lock(filp, file_lock, &flock);
1892 if (cmd == F_SETLKW) {
1893 file_lock->fl_flags |= FL_SLEEP;
1897 switch (flock.l_type) {
1899 if (!(filp->f_mode & FMODE_READ))
1903 if (!(filp->f_mode & FMODE_WRITE))
1913 error = do_lock_file_wait(filp, cmd, file_lock);
1916 * Attempt to detect a close/fcntl race and recover by
1917 * releasing the lock that was just acquired.
1920 * we need that spin_lock here - it prevents reordering between
1921 * update of inode->i_flock and check for it done in close().
1922 * rcu_read_lock() wouldn't do.
1924 spin_lock(¤t->files->file_lock);
1926 spin_unlock(¤t->files->file_lock);
1927 if (!error && f != filp && flock.l_type != F_UNLCK) {
1928 flock.l_type = F_UNLCK;
1933 locks_free_lock(file_lock);
1937 #if BITS_PER_LONG == 32
1938 /* Report the first existing lock that would conflict with l.
1939 * This implements the F_GETLK command of fcntl().
1941 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1943 struct file_lock file_lock;
1944 struct flock64 flock;
1948 if (copy_from_user(&flock, l, sizeof(flock)))
1951 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1954 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1958 error = vfs_test_lock(filp, &file_lock);
1962 flock.l_type = file_lock.fl_type;
1963 if (file_lock.fl_type != F_UNLCK)
1964 posix_lock_to_flock64(&flock, &file_lock);
1967 if (!copy_to_user(l, &flock, sizeof(flock)))
1974 /* Apply the lock described by l to an open file descriptor.
1975 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1977 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1978 struct flock64 __user *l)
1980 struct file_lock *file_lock = locks_alloc_lock();
1981 struct flock64 flock;
1982 struct inode *inode;
1986 if (file_lock == NULL)
1990 * This might block, so we do it before checking the inode.
1993 if (copy_from_user(&flock, l, sizeof(flock)))
1996 inode = file_inode(filp);
1998 /* Don't allow mandatory locks on files that may be memory mapped
2001 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2007 error = flock64_to_posix_lock(filp, file_lock, &flock);
2010 if (cmd == F_SETLKW64) {
2011 file_lock->fl_flags |= FL_SLEEP;
2015 switch (flock.l_type) {
2017 if (!(filp->f_mode & FMODE_READ))
2021 if (!(filp->f_mode & FMODE_WRITE))
2031 error = do_lock_file_wait(filp, cmd, file_lock);
2034 * Attempt to detect a close/fcntl race and recover by
2035 * releasing the lock that was just acquired.
2037 spin_lock(¤t->files->file_lock);
2039 spin_unlock(¤t->files->file_lock);
2040 if (!error && f != filp && flock.l_type != F_UNLCK) {
2041 flock.l_type = F_UNLCK;
2046 locks_free_lock(file_lock);
2049 #endif /* BITS_PER_LONG == 32 */
2052 * This function is called when the file is being removed
2053 * from the task's fd array. POSIX locks belonging to this task
2054 * are deleted at this time.
2056 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2058 struct file_lock lock;
2061 * If there are no locks held on this file, we don't need to call
2062 * posix_lock_file(). Another process could be setting a lock on this
2063 * file at the same time, but we wouldn't remove that lock anyway.
2065 if (!file_inode(filp)->i_flock)
2068 lock.fl_type = F_UNLCK;
2069 lock.fl_flags = FL_POSIX | FL_CLOSE;
2071 lock.fl_end = OFFSET_MAX;
2072 lock.fl_owner = owner;
2073 lock.fl_pid = current->tgid;
2074 lock.fl_file = filp;
2076 lock.fl_lmops = NULL;
2078 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2080 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2081 lock.fl_ops->fl_release_private(&lock);
2084 EXPORT_SYMBOL(locks_remove_posix);
2087 * This function is called on the last close of an open file.
2089 void locks_remove_flock(struct file *filp)
2091 struct inode * inode = file_inode(filp);
2092 struct file_lock *fl;
2093 struct file_lock **before;
2095 if (!inode->i_flock)
2098 if (filp->f_op && filp->f_op->flock) {
2099 struct file_lock fl = {
2100 .fl_pid = current->tgid,
2102 .fl_flags = FL_FLOCK,
2104 .fl_end = OFFSET_MAX,
2106 filp->f_op->flock(filp, F_SETLKW, &fl);
2107 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2108 fl.fl_ops->fl_release_private(&fl);
2112 before = &inode->i_flock;
2114 while ((fl = *before) != NULL) {
2115 if (fl->fl_file == filp) {
2117 locks_delete_lock(before);
2121 lease_modify(before, F_UNLCK);
2127 before = &fl->fl_next;
2133 * posix_unblock_lock - stop waiting for a file lock
2134 * @waiter: the lock which was waiting
2136 * lockd needs to block waiting for locks.
2139 posix_unblock_lock(struct file_lock *waiter)
2144 if (waiter->fl_next)
2145 __locks_delete_block(waiter);
2151 EXPORT_SYMBOL(posix_unblock_lock);
2154 * vfs_cancel_lock - file byte range unblock lock
2155 * @filp: The file to apply the unblock to
2156 * @fl: The lock to be unblocked
2158 * Used by lock managers to cancel blocked requests
2160 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2162 if (filp->f_op && filp->f_op->lock)
2163 return filp->f_op->lock(filp, F_CANCELLK, fl);
2167 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2169 #ifdef CONFIG_PROC_FS
2170 #include <linux/proc_fs.h>
2171 #include <linux/seq_file.h>
2173 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2174 loff_t id, char *pfx)
2176 struct inode *inode = NULL;
2177 unsigned int fl_pid;
2180 fl_pid = pid_vnr(fl->fl_nspid);
2182 fl_pid = fl->fl_pid;
2184 if (fl->fl_file != NULL)
2185 inode = file_inode(fl->fl_file);
2187 seq_printf(f, "%lld:%s ", id, pfx);
2189 seq_printf(f, "%6s %s ",
2190 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2191 (inode == NULL) ? "*NOINODE*" :
2192 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2193 } else if (IS_FLOCK(fl)) {
2194 if (fl->fl_type & LOCK_MAND) {
2195 seq_printf(f, "FLOCK MSNFS ");
2197 seq_printf(f, "FLOCK ADVISORY ");
2199 } else if (IS_LEASE(fl)) {
2200 seq_printf(f, "LEASE ");
2201 if (lease_breaking(fl))
2202 seq_printf(f, "BREAKING ");
2203 else if (fl->fl_file)
2204 seq_printf(f, "ACTIVE ");
2206 seq_printf(f, "BREAKER ");
2208 seq_printf(f, "UNKNOWN UNKNOWN ");
2210 if (fl->fl_type & LOCK_MAND) {
2211 seq_printf(f, "%s ",
2212 (fl->fl_type & LOCK_READ)
2213 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2214 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2216 seq_printf(f, "%s ",
2217 (lease_breaking(fl))
2218 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2219 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2222 #ifdef WE_CAN_BREAK_LSLK_NOW
2223 seq_printf(f, "%d %s:%ld ", fl_pid,
2224 inode->i_sb->s_id, inode->i_ino);
2226 /* userspace relies on this representation of dev_t ;-( */
2227 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2228 MAJOR(inode->i_sb->s_dev),
2229 MINOR(inode->i_sb->s_dev), inode->i_ino);
2232 seq_printf(f, "%d <none>:0 ", fl_pid);
2235 if (fl->fl_end == OFFSET_MAX)
2236 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2238 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2240 seq_printf(f, "0 EOF\n");
2244 static int locks_show(struct seq_file *f, void *v)
2246 struct file_lock *fl, *bfl;
2248 fl = list_entry(v, struct file_lock, fl_link);
2250 lock_get_status(f, fl, *((loff_t *)f->private), "");
2252 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2253 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2258 static void *locks_start(struct seq_file *f, loff_t *pos)
2260 loff_t *p = f->private;
2264 return seq_list_start(&file_lock_list, *pos);
2267 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2269 loff_t *p = f->private;
2271 return seq_list_next(v, &file_lock_list, pos);
2274 static void locks_stop(struct seq_file *f, void *v)
2279 static const struct seq_operations locks_seq_operations = {
2280 .start = locks_start,
2286 static int locks_open(struct inode *inode, struct file *filp)
2288 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2291 static const struct file_operations proc_locks_operations = {
2294 .llseek = seq_lseek,
2295 .release = seq_release_private,
2298 static int __init proc_locks_init(void)
2300 proc_create("locks", 0, NULL, &proc_locks_operations);
2303 module_init(proc_locks_init);
2307 * lock_may_read - checks that the region is free of locks
2308 * @inode: the inode that is being read
2309 * @start: the first byte to read
2310 * @len: the number of bytes to read
2312 * Emulates Windows locking requirements. Whole-file
2313 * mandatory locks (share modes) can prohibit a read and
2314 * byte-range POSIX locks can prohibit a read if they overlap.
2316 * N.B. this function is only ever called
2317 * from knfsd and ownership of locks is never checked.
2319 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2321 struct file_lock *fl;
2324 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2326 if (fl->fl_type == F_RDLCK)
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_READ)
2344 EXPORT_SYMBOL(lock_may_read);
2347 * lock_may_write - checks that the region is free of locks
2348 * @inode: the inode that is being written
2349 * @start: the first byte to write
2350 * @len: the number of bytes to write
2352 * Emulates Windows locking requirements. Whole-file
2353 * mandatory locks (share modes) can prohibit a write and
2354 * byte-range POSIX locks can prohibit a write if they overlap.
2356 * N.B. this function is only ever called
2357 * from knfsd and ownership of locks is never checked.
2359 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2361 struct file_lock *fl;
2364 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2366 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2368 } else if (IS_FLOCK(fl)) {
2369 if (!(fl->fl_type & LOCK_MAND))
2371 if (fl->fl_type & LOCK_WRITE)
2382 EXPORT_SYMBOL(lock_may_write);
2384 static int __init filelock_init(void)
2386 filelock_cache = kmem_cache_create("file_lock_cache",
2387 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2392 core_initcall(filelock_init);