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 static LIST_HEAD(file_lock_list);
157 static LIST_HEAD(blocked_list);
158 static DEFINE_SPINLOCK(file_lock_lock);
161 * Protects the two list heads above, plus the inode->i_flock list
163 void lock_flocks(void)
165 spin_lock(&file_lock_lock);
167 EXPORT_SYMBOL_GPL(lock_flocks);
169 void unlock_flocks(void)
171 spin_unlock(&file_lock_lock);
173 EXPORT_SYMBOL_GPL(unlock_flocks);
175 static struct kmem_cache *filelock_cache __read_mostly;
177 static void locks_init_lock_heads(struct file_lock *fl)
179 INIT_LIST_HEAD(&fl->fl_link);
180 INIT_LIST_HEAD(&fl->fl_block);
181 init_waitqueue_head(&fl->fl_wait);
184 /* Allocate an empty lock structure. */
185 struct file_lock *locks_alloc_lock(void)
187 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
190 locks_init_lock_heads(fl);
194 EXPORT_SYMBOL_GPL(locks_alloc_lock);
196 void locks_release_private(struct file_lock *fl)
199 if (fl->fl_ops->fl_release_private)
200 fl->fl_ops->fl_release_private(fl);
206 EXPORT_SYMBOL_GPL(locks_release_private);
208 /* Free a lock which is not in use. */
209 void locks_free_lock(struct file_lock *fl)
211 BUG_ON(waitqueue_active(&fl->fl_wait));
212 BUG_ON(!list_empty(&fl->fl_block));
213 BUG_ON(!list_empty(&fl->fl_link));
215 locks_release_private(fl);
216 kmem_cache_free(filelock_cache, fl);
218 EXPORT_SYMBOL(locks_free_lock);
220 void locks_init_lock(struct file_lock *fl)
222 memset(fl, 0, sizeof(struct file_lock));
223 locks_init_lock_heads(fl);
226 EXPORT_SYMBOL(locks_init_lock);
228 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
231 if (fl->fl_ops->fl_copy_lock)
232 fl->fl_ops->fl_copy_lock(new, fl);
233 new->fl_ops = fl->fl_ops;
236 new->fl_lmops = fl->fl_lmops;
240 * Initialize a new lock from an existing file_lock structure.
242 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
244 new->fl_owner = fl->fl_owner;
245 new->fl_pid = fl->fl_pid;
247 new->fl_flags = fl->fl_flags;
248 new->fl_type = fl->fl_type;
249 new->fl_start = fl->fl_start;
250 new->fl_end = fl->fl_end;
252 new->fl_lmops = NULL;
254 EXPORT_SYMBOL(__locks_copy_lock);
256 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
258 locks_release_private(new);
260 __locks_copy_lock(new, fl);
261 new->fl_file = fl->fl_file;
262 new->fl_ops = fl->fl_ops;
263 new->fl_lmops = fl->fl_lmops;
265 locks_copy_private(new, fl);
268 EXPORT_SYMBOL(locks_copy_lock);
270 static inline int flock_translate_cmd(int cmd) {
272 return cmd & (LOCK_MAND | LOCK_RW);
284 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
285 static int flock_make_lock(struct file *filp, struct file_lock **lock,
288 struct file_lock *fl;
289 int type = flock_translate_cmd(cmd);
293 fl = locks_alloc_lock();
298 fl->fl_pid = current->tgid;
299 fl->fl_flags = FL_FLOCK;
301 fl->fl_end = OFFSET_MAX;
307 static int assign_type(struct file_lock *fl, long type)
321 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
324 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
329 switch (l->l_whence) {
337 start = i_size_read(file_inode(filp));
343 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
344 POSIX-2001 defines it. */
348 fl->fl_end = OFFSET_MAX;
350 end = start + l->l_len - 1;
352 } else if (l->l_len < 0) {
359 fl->fl_start = start; /* we record the absolute position */
360 if (fl->fl_end < fl->fl_start)
363 fl->fl_owner = current->files;
364 fl->fl_pid = current->tgid;
366 fl->fl_flags = FL_POSIX;
370 return assign_type(fl, l->l_type);
373 #if BITS_PER_LONG == 32
374 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
379 switch (l->l_whence) {
387 start = i_size_read(file_inode(filp));
396 fl->fl_end = OFFSET_MAX;
398 fl->fl_end = start + l->l_len - 1;
399 } else if (l->l_len < 0) {
400 fl->fl_end = start - 1;
405 fl->fl_start = start; /* we record the absolute position */
406 if (fl->fl_end < fl->fl_start)
409 fl->fl_owner = current->files;
410 fl->fl_pid = current->tgid;
412 fl->fl_flags = FL_POSIX;
416 return assign_type(fl, l->l_type);
420 /* default lease lock manager operations */
421 static void lease_break_callback(struct file_lock *fl)
423 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
426 static const struct lock_manager_operations lease_manager_ops = {
427 .lm_break = lease_break_callback,
428 .lm_change = lease_modify,
432 * Initialize a lease, use the default lock manager operations
434 static int lease_init(struct file *filp, long type, struct file_lock *fl)
436 if (assign_type(fl, type) != 0)
439 fl->fl_owner = current->files;
440 fl->fl_pid = current->tgid;
443 fl->fl_flags = FL_LEASE;
445 fl->fl_end = OFFSET_MAX;
447 fl->fl_lmops = &lease_manager_ops;
451 /* Allocate a file_lock initialised to this type of lease */
452 static struct file_lock *lease_alloc(struct file *filp, long type)
454 struct file_lock *fl = locks_alloc_lock();
458 return ERR_PTR(error);
460 error = lease_init(filp, type, fl);
463 return ERR_PTR(error);
468 /* Check if two locks overlap each other.
470 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
472 return ((fl1->fl_end >= fl2->fl_start) &&
473 (fl2->fl_end >= fl1->fl_start));
477 * Check whether two locks have the same owner.
479 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
481 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
482 return fl2->fl_lmops == fl1->fl_lmops &&
483 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
484 return fl1->fl_owner == fl2->fl_owner;
487 /* Remove waiter from blocker's block list.
488 * When blocker ends up pointing to itself then the list is empty.
490 static void __locks_delete_block(struct file_lock *waiter)
492 list_del_init(&waiter->fl_block);
493 list_del_init(&waiter->fl_link);
494 waiter->fl_next = NULL;
499 static void locks_delete_block(struct file_lock *waiter)
502 __locks_delete_block(waiter);
506 /* Insert waiter into blocker's block list.
507 * We use a circular list so that processes can be easily woken up in
508 * the order they blocked. The documentation doesn't require this but
509 * it seems like the reasonable thing to do.
511 static void locks_insert_block(struct file_lock *blocker,
512 struct file_lock *waiter)
514 BUG_ON(!list_empty(&waiter->fl_block));
515 list_add_tail(&waiter->fl_block, &blocker->fl_block);
516 waiter->fl_next = blocker;
517 if (IS_POSIX(blocker))
518 list_add(&waiter->fl_link, &blocked_list);
522 * Wake up processes blocked waiting for blocker.
524 * Must be called with the file_lock_lock held!
526 static void locks_wake_up_blocks(struct file_lock *blocker)
528 while (!list_empty(&blocker->fl_block)) {
529 struct file_lock *waiter;
531 waiter = list_first_entry(&blocker->fl_block,
532 struct file_lock, fl_block);
533 __locks_delete_block(waiter);
534 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
535 waiter->fl_lmops->lm_notify(waiter);
537 wake_up(&waiter->fl_wait);
541 /* Insert file lock fl into an inode's lock list at the position indicated
542 * by pos. At the same time add the lock to the global file lock list.
544 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
546 list_add(&fl->fl_link, &file_lock_list);
548 fl->fl_nspid = get_pid(task_tgid(current));
550 /* insert into file's list */
556 * Delete a lock and then free it.
557 * Wake up processes that are blocked waiting for this lock,
558 * notify the FS that the lock has been cleared and
559 * finally free the lock.
561 static void locks_delete_lock(struct file_lock **thisfl_p)
563 struct file_lock *fl = *thisfl_p;
565 *thisfl_p = fl->fl_next;
567 list_del_init(&fl->fl_link);
570 put_pid(fl->fl_nspid);
574 locks_wake_up_blocks(fl);
578 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
579 * checks for shared/exclusive status of overlapping locks.
581 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
583 if (sys_fl->fl_type == F_WRLCK)
585 if (caller_fl->fl_type == F_WRLCK)
590 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
591 * checking before calling the locks_conflict().
593 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
595 /* POSIX locks owned by the same process do not conflict with
598 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
601 /* Check whether they overlap */
602 if (!locks_overlap(caller_fl, sys_fl))
605 return (locks_conflict(caller_fl, sys_fl));
608 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
609 * checking before calling the locks_conflict().
611 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
613 /* FLOCK locks referring to the same filp do not conflict with
616 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
618 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
621 return (locks_conflict(caller_fl, sys_fl));
625 posix_test_lock(struct file *filp, struct file_lock *fl)
627 struct file_lock *cfl;
630 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
633 if (posix_locks_conflict(fl, cfl))
637 __locks_copy_lock(fl, cfl);
639 fl->fl_pid = pid_vnr(cfl->fl_nspid);
641 fl->fl_type = F_UNLCK;
645 EXPORT_SYMBOL(posix_test_lock);
648 * Deadlock detection:
650 * We attempt to detect deadlocks that are due purely to posix file
653 * We assume that a task can be waiting for at most one lock at a time.
654 * So for any acquired lock, the process holding that lock may be
655 * waiting on at most one other lock. That lock in turns may be held by
656 * someone waiting for at most one other lock. Given a requested lock
657 * caller_fl which is about to wait for a conflicting lock block_fl, we
658 * follow this chain of waiters to ensure we are not about to create a
661 * Since we do this before we ever put a process to sleep on a lock, we
662 * are ensured that there is never a cycle; that is what guarantees that
663 * the while() loop in posix_locks_deadlock() eventually completes.
665 * Note: the above assumption may not be true when handling lock
666 * requests from a broken NFS client. It may also fail in the presence
667 * of tasks (such as posix threads) sharing the same open file table.
669 * To handle those cases, we just bail out after a few iterations.
672 #define MAX_DEADLK_ITERATIONS 10
674 /* Find a lock that the owner of the given block_fl is blocking on. */
675 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
677 struct file_lock *fl;
679 list_for_each_entry(fl, &blocked_list, fl_link) {
680 if (posix_same_owner(fl, block_fl))
686 static int posix_locks_deadlock(struct file_lock *caller_fl,
687 struct file_lock *block_fl)
691 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
692 if (i++ > MAX_DEADLK_ITERATIONS)
694 if (posix_same_owner(caller_fl, block_fl))
700 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
701 * after any leases, but before any posix locks.
703 * Note that if called with an FL_EXISTS argument, the caller may determine
704 * whether or not a lock was successfully freed by testing the return
707 static int flock_lock_file(struct file *filp, struct file_lock *request)
709 struct file_lock *new_fl = NULL;
710 struct file_lock **before;
711 struct inode * inode = file_inode(filp);
715 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
716 new_fl = locks_alloc_lock();
722 if (request->fl_flags & FL_ACCESS)
725 for_each_lock(inode, before) {
726 struct file_lock *fl = *before;
731 if (filp != fl->fl_file)
733 if (request->fl_type == fl->fl_type)
736 locks_delete_lock(before);
740 if (request->fl_type == F_UNLCK) {
741 if ((request->fl_flags & FL_EXISTS) && !found)
747 * If a higher-priority process was blocked on the old file lock,
748 * give it the opportunity to lock the file.
757 for_each_lock(inode, before) {
758 struct file_lock *fl = *before;
763 if (!flock_locks_conflict(request, fl))
766 if (!(request->fl_flags & FL_SLEEP))
768 error = FILE_LOCK_DEFERRED;
769 locks_insert_block(fl, request);
772 if (request->fl_flags & FL_ACCESS)
774 locks_copy_lock(new_fl, request);
775 locks_insert_lock(before, new_fl);
782 locks_free_lock(new_fl);
786 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
788 struct file_lock *fl;
789 struct file_lock *new_fl = NULL;
790 struct file_lock *new_fl2 = NULL;
791 struct file_lock *left = NULL;
792 struct file_lock *right = NULL;
793 struct file_lock **before;
794 int error, added = 0;
797 * We may need two file_lock structures for this operation,
798 * so we get them in advance to avoid races.
800 * In some cases we can be sure, that no new locks will be needed
802 if (!(request->fl_flags & FL_ACCESS) &&
803 (request->fl_type != F_UNLCK ||
804 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
811 * New lock request. Walk all POSIX locks and look for conflicts. If
812 * there are any, either return error or put the request on the
813 * blocker's list of waiters and the global blocked_list.
815 if (request->fl_type != F_UNLCK) {
816 for_each_lock(inode, before) {
820 if (!posix_locks_conflict(request, fl))
823 __locks_copy_lock(conflock, fl);
825 if (!(request->fl_flags & FL_SLEEP))
828 if (posix_locks_deadlock(request, fl))
830 error = FILE_LOCK_DEFERRED;
831 locks_insert_block(fl, request);
836 /* If we're just looking for a conflict, we're done. */
838 if (request->fl_flags & FL_ACCESS)
842 * Find the first old lock with the same owner as the new lock.
845 before = &inode->i_flock;
847 /* First skip locks owned by other processes. */
848 while ((fl = *before) && (!IS_POSIX(fl) ||
849 !posix_same_owner(request, fl))) {
850 before = &fl->fl_next;
853 /* Process locks with this owner. */
854 while ((fl = *before) && posix_same_owner(request, fl)) {
855 /* Detect adjacent or overlapping regions (if same lock type)
857 if (request->fl_type == fl->fl_type) {
858 /* In all comparisons of start vs end, use
859 * "start - 1" rather than "end + 1". If end
860 * is OFFSET_MAX, end + 1 will become negative.
862 if (fl->fl_end < request->fl_start - 1)
864 /* If the next lock in the list has entirely bigger
865 * addresses than the new one, insert the lock here.
867 if (fl->fl_start - 1 > request->fl_end)
870 /* If we come here, the new and old lock are of the
871 * same type and adjacent or overlapping. Make one
872 * lock yielding from the lower start address of both
873 * locks to the higher end address.
875 if (fl->fl_start > request->fl_start)
876 fl->fl_start = request->fl_start;
878 request->fl_start = fl->fl_start;
879 if (fl->fl_end < request->fl_end)
880 fl->fl_end = request->fl_end;
882 request->fl_end = fl->fl_end;
884 locks_delete_lock(before);
891 /* Processing for different lock types is a bit
894 if (fl->fl_end < request->fl_start)
896 if (fl->fl_start > request->fl_end)
898 if (request->fl_type == F_UNLCK)
900 if (fl->fl_start < request->fl_start)
902 /* If the next lock in the list has a higher end
903 * address than the new one, insert the new one here.
905 if (fl->fl_end > request->fl_end) {
909 if (fl->fl_start >= request->fl_start) {
910 /* The new lock completely replaces an old
911 * one (This may happen several times).
914 locks_delete_lock(before);
917 /* Replace the old lock with the new one.
918 * Wake up anybody waiting for the old one,
919 * as the change in lock type might satisfy
922 locks_wake_up_blocks(fl);
923 fl->fl_start = request->fl_start;
924 fl->fl_end = request->fl_end;
925 fl->fl_type = request->fl_type;
926 locks_release_private(fl);
927 locks_copy_private(fl, request);
932 /* Go on to next lock.
935 before = &fl->fl_next;
939 * The above code only modifies existing locks in case of merging or
940 * replacing. If new lock(s) need to be inserted all modifications are
941 * done below this, so it's safe yet to bail out.
943 error = -ENOLCK; /* "no luck" */
944 if (right && left == right && !new_fl2)
949 if (request->fl_type == F_UNLCK) {
950 if (request->fl_flags & FL_EXISTS)
959 locks_copy_lock(new_fl, request);
960 locks_insert_lock(before, new_fl);
965 /* The new lock breaks the old one in two pieces,
966 * so we have to use the second new lock.
970 locks_copy_lock(left, right);
971 locks_insert_lock(before, left);
973 right->fl_start = request->fl_end + 1;
974 locks_wake_up_blocks(right);
977 left->fl_end = request->fl_start - 1;
978 locks_wake_up_blocks(left);
983 * Free any unused locks.
986 locks_free_lock(new_fl);
988 locks_free_lock(new_fl2);
993 * posix_lock_file - Apply a POSIX-style lock to a file
994 * @filp: The file to apply the lock to
995 * @fl: The lock to be applied
996 * @conflock: Place to return a copy of the conflicting lock, if found.
998 * Add a POSIX style lock to a file.
999 * We merge adjacent & overlapping locks whenever possible.
1000 * POSIX locks are sorted by owner task, then by starting address
1002 * Note that if called with an FL_EXISTS argument, the caller may determine
1003 * whether or not a lock was successfully freed by testing the return
1004 * value for -ENOENT.
1006 int posix_lock_file(struct file *filp, struct file_lock *fl,
1007 struct file_lock *conflock)
1009 return __posix_lock_file(file_inode(filp), fl, conflock);
1011 EXPORT_SYMBOL(posix_lock_file);
1014 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1015 * @filp: The file to apply the lock to
1016 * @fl: The lock to be applied
1018 * Add a POSIX style lock to a file.
1019 * We merge adjacent & overlapping locks whenever possible.
1020 * POSIX locks are sorted by owner task, then by starting address
1022 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1027 error = posix_lock_file(filp, fl, NULL);
1028 if (error != FILE_LOCK_DEFERRED)
1030 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1034 locks_delete_block(fl);
1039 EXPORT_SYMBOL(posix_lock_file_wait);
1042 * locks_mandatory_locked - Check for an active lock
1043 * @inode: the file to check
1045 * Searches the inode's list of locks to find any POSIX locks which conflict.
1046 * This function is called from locks_verify_locked() only.
1048 int locks_mandatory_locked(struct inode *inode)
1050 fl_owner_t owner = current->files;
1051 struct file_lock *fl;
1054 * Search the lock list for this inode for any POSIX locks.
1057 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1060 if (fl->fl_owner != owner)
1064 return fl ? -EAGAIN : 0;
1068 * locks_mandatory_area - Check for a conflicting lock
1069 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1071 * @inode: the file to check
1072 * @filp: how the file was opened (if it was)
1073 * @offset: start of area to check
1074 * @count: length of area to check
1076 * Searches the inode's list of locks to find any POSIX locks which conflict.
1077 * This function is called from rw_verify_area() and
1078 * locks_verify_truncate().
1080 int locks_mandatory_area(int read_write, struct inode *inode,
1081 struct file *filp, loff_t offset,
1084 struct file_lock fl;
1087 locks_init_lock(&fl);
1088 fl.fl_owner = current->files;
1089 fl.fl_pid = current->tgid;
1091 fl.fl_flags = FL_POSIX | FL_ACCESS;
1092 if (filp && !(filp->f_flags & O_NONBLOCK))
1093 fl.fl_flags |= FL_SLEEP;
1094 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1095 fl.fl_start = offset;
1096 fl.fl_end = offset + count - 1;
1099 error = __posix_lock_file(inode, &fl, NULL);
1100 if (error != FILE_LOCK_DEFERRED)
1102 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1105 * If we've been sleeping someone might have
1106 * changed the permissions behind our back.
1108 if (__mandatory_lock(inode))
1112 locks_delete_block(&fl);
1119 EXPORT_SYMBOL(locks_mandatory_area);
1121 static void lease_clear_pending(struct file_lock *fl, int arg)
1125 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1128 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1132 /* We already had a lease on this file; just change its type */
1133 int lease_modify(struct file_lock **before, int arg)
1135 struct file_lock *fl = *before;
1136 int error = assign_type(fl, arg);
1140 lease_clear_pending(fl, arg);
1141 locks_wake_up_blocks(fl);
1142 if (arg == F_UNLCK) {
1143 struct file *filp = fl->fl_file;
1146 filp->f_owner.signum = 0;
1147 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1148 if (fl->fl_fasync != NULL) {
1149 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1150 fl->fl_fasync = NULL;
1152 locks_delete_lock(before);
1157 EXPORT_SYMBOL(lease_modify);
1159 static bool past_time(unsigned long then)
1162 /* 0 is a special value meaning "this never expires": */
1164 return time_after(jiffies, then);
1167 static void time_out_leases(struct inode *inode)
1169 struct file_lock **before;
1170 struct file_lock *fl;
1172 before = &inode->i_flock;
1173 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1174 if (past_time(fl->fl_downgrade_time))
1175 lease_modify(before, F_RDLCK);
1176 if (past_time(fl->fl_break_time))
1177 lease_modify(before, F_UNLCK);
1178 if (fl == *before) /* lease_modify may have freed fl */
1179 before = &fl->fl_next;
1184 * __break_lease - revoke all outstanding leases on file
1185 * @inode: the inode of the file to return
1186 * @mode: the open mode (read or write)
1188 * break_lease (inlined for speed) has checked there already is at least
1189 * some kind of lock (maybe a lease) on this file. Leases are broken on
1190 * a call to open() or truncate(). This function can sleep unless you
1191 * specified %O_NONBLOCK to your open().
1193 int __break_lease(struct inode *inode, unsigned int mode)
1196 struct file_lock *new_fl, *flock;
1197 struct file_lock *fl;
1198 unsigned long break_time;
1199 int i_have_this_lease = 0;
1200 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1202 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1204 return PTR_ERR(new_fl);
1208 time_out_leases(inode);
1210 flock = inode->i_flock;
1211 if ((flock == NULL) || !IS_LEASE(flock))
1214 if (!locks_conflict(flock, new_fl))
1217 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1218 if (fl->fl_owner == current->files)
1219 i_have_this_lease = 1;
1222 if (lease_break_time > 0) {
1223 break_time = jiffies + lease_break_time * HZ;
1224 if (break_time == 0)
1225 break_time++; /* so that 0 means no break time */
1228 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1230 if (fl->fl_flags & FL_UNLOCK_PENDING)
1232 fl->fl_flags |= FL_UNLOCK_PENDING;
1233 fl->fl_break_time = break_time;
1235 if (lease_breaking(flock))
1237 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1238 fl->fl_downgrade_time = break_time;
1240 fl->fl_lmops->lm_break(fl);
1243 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1244 error = -EWOULDBLOCK;
1249 break_time = flock->fl_break_time;
1250 if (break_time != 0) {
1251 break_time -= jiffies;
1252 if (break_time == 0)
1255 locks_insert_block(flock, new_fl);
1257 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1258 !new_fl->fl_next, break_time);
1260 __locks_delete_block(new_fl);
1263 time_out_leases(inode);
1265 * Wait for the next conflicting lease that has not been
1268 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1269 flock = flock->fl_next) {
1270 if (locks_conflict(new_fl, flock))
1278 locks_free_lock(new_fl);
1282 EXPORT_SYMBOL(__break_lease);
1285 * lease_get_mtime - get the last modified time of an inode
1287 * @time: pointer to a timespec which will contain the last modified time
1289 * This is to force NFS clients to flush their caches for files with
1290 * exclusive leases. The justification is that if someone has an
1291 * exclusive lease, then they could be modifying it.
1293 void lease_get_mtime(struct inode *inode, struct timespec *time)
1295 struct file_lock *flock = inode->i_flock;
1296 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1297 *time = current_fs_time(inode->i_sb);
1299 *time = inode->i_mtime;
1302 EXPORT_SYMBOL(lease_get_mtime);
1305 * fcntl_getlease - Enquire what lease is currently active
1308 * The value returned by this function will be one of
1309 * (if no lease break is pending):
1311 * %F_RDLCK to indicate a shared lease is held.
1313 * %F_WRLCK to indicate an exclusive lease is held.
1315 * %F_UNLCK to indicate no lease is held.
1317 * (if a lease break is pending):
1319 * %F_RDLCK to indicate an exclusive lease needs to be
1320 * changed to a shared lease (or removed).
1322 * %F_UNLCK to indicate the lease needs to be removed.
1324 * XXX: sfr & willy disagree over whether F_INPROGRESS
1325 * should be returned to userspace.
1327 int fcntl_getlease(struct file *filp)
1329 struct file_lock *fl;
1333 time_out_leases(file_inode(filp));
1334 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1336 if (fl->fl_file == filp) {
1337 type = target_leasetype(fl);
1345 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1347 struct file_lock *fl, **before, **my_before = NULL, *lease;
1348 struct dentry *dentry = filp->f_path.dentry;
1349 struct inode *inode = dentry->d_inode;
1355 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1357 if ((arg == F_WRLCK)
1358 && ((dentry->d_count > 1)
1359 || (atomic_read(&inode->i_count) > 1)))
1363 * At this point, we know that if there is an exclusive
1364 * lease on this file, then we hold it on this filp
1365 * (otherwise our open of this file would have blocked).
1366 * And if we are trying to acquire an exclusive lease,
1367 * then the file is not open by anyone (including us)
1368 * except for this filp.
1371 for (before = &inode->i_flock;
1372 ((fl = *before) != NULL) && IS_LEASE(fl);
1373 before = &fl->fl_next) {
1374 if (fl->fl_file == filp) {
1379 * No exclusive leases if someone else has a lease on
1385 * Modifying our existing lease is OK, but no getting a
1386 * new lease if someone else is opening for write:
1388 if (fl->fl_flags & FL_UNLOCK_PENDING)
1392 if (my_before != NULL) {
1393 error = lease->fl_lmops->lm_change(my_before, arg);
1403 locks_insert_lock(before, lease);
1410 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1412 struct file_lock *fl, **before;
1413 struct dentry *dentry = filp->f_path.dentry;
1414 struct inode *inode = dentry->d_inode;
1416 for (before = &inode->i_flock;
1417 ((fl = *before) != NULL) && IS_LEASE(fl);
1418 before = &fl->fl_next) {
1419 if (fl->fl_file != filp)
1421 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1427 * generic_setlease - sets a lease on an open file
1428 * @filp: file pointer
1429 * @arg: type of lease to obtain
1430 * @flp: input - file_lock to use, output - file_lock inserted
1432 * The (input) flp->fl_lmops->lm_break function is required
1435 * Called with file_lock_lock held.
1437 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1439 struct dentry *dentry = filp->f_path.dentry;
1440 struct inode *inode = dentry->d_inode;
1443 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1445 if (!S_ISREG(inode->i_mode))
1447 error = security_file_lock(filp, arg);
1451 time_out_leases(inode);
1453 BUG_ON(!(*flp)->fl_lmops->lm_break);
1457 return generic_delete_lease(filp, flp);
1460 return generic_add_lease(filp, arg, flp);
1465 EXPORT_SYMBOL(generic_setlease);
1467 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1469 if (filp->f_op && filp->f_op->setlease)
1470 return filp->f_op->setlease(filp, arg, lease);
1472 return generic_setlease(filp, arg, lease);
1476 * vfs_setlease - sets a lease on an open file
1477 * @filp: file pointer
1478 * @arg: type of lease to obtain
1479 * @lease: file_lock to use
1481 * Call this to establish a lease on the file.
1482 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1483 * break_lease will oops!
1485 * This will call the filesystem's setlease file method, if
1486 * defined. Note that there is no getlease method; instead, the
1487 * filesystem setlease method should call back to setlease() to
1488 * add a lease to the inode's lease list, where fcntl_getlease() can
1489 * find it. Since fcntl_getlease() only reports whether the current
1490 * task holds a lease, a cluster filesystem need only do this for
1491 * leases held by processes on this node.
1493 * There is also no break_lease method; filesystems that
1494 * handle their own leases should break leases themselves from the
1495 * filesystem's open, create, and (on truncate) setattr methods.
1497 * Warning: the only current setlease methods exist only to disable
1498 * leases in certain cases. More vfs changes may be required to
1499 * allow a full filesystem lease implementation.
1502 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1507 error = __vfs_setlease(filp, arg, lease);
1512 EXPORT_SYMBOL_GPL(vfs_setlease);
1514 static int do_fcntl_delete_lease(struct file *filp)
1516 struct file_lock fl, *flp = &fl;
1518 lease_init(filp, F_UNLCK, flp);
1520 return vfs_setlease(filp, F_UNLCK, &flp);
1523 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1525 struct file_lock *fl, *ret;
1526 struct fasync_struct *new;
1529 fl = lease_alloc(filp, arg);
1533 new = fasync_alloc();
1535 locks_free_lock(fl);
1540 error = __vfs_setlease(filp, arg, &ret);
1543 locks_free_lock(fl);
1544 goto out_free_fasync;
1547 locks_free_lock(fl);
1550 * fasync_insert_entry() returns the old entry if any.
1551 * If there was no old entry, then it used 'new' and
1552 * inserted it into the fasync list. Clear new so that
1553 * we don't release it here.
1555 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1558 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1568 * fcntl_setlease - sets a lease on an open file
1569 * @fd: open file descriptor
1570 * @filp: file pointer
1571 * @arg: type of lease to obtain
1573 * Call this fcntl to establish a lease on the file.
1574 * Note that you also need to call %F_SETSIG to
1575 * receive a signal when the lease is broken.
1577 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1580 return do_fcntl_delete_lease(filp);
1581 return do_fcntl_add_lease(fd, filp, arg);
1585 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1586 * @filp: The file to apply the lock to
1587 * @fl: The lock to be applied
1589 * Add a FLOCK style lock to a file.
1591 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1596 error = flock_lock_file(filp, fl);
1597 if (error != FILE_LOCK_DEFERRED)
1599 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1603 locks_delete_block(fl);
1609 EXPORT_SYMBOL(flock_lock_file_wait);
1612 * sys_flock: - flock() system call.
1613 * @fd: the file descriptor to lock.
1614 * @cmd: the type of lock to apply.
1616 * Apply a %FL_FLOCK style lock to an open file descriptor.
1617 * The @cmd can be one of
1619 * %LOCK_SH -- a shared lock.
1621 * %LOCK_EX -- an exclusive lock.
1623 * %LOCK_UN -- remove an existing lock.
1625 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1627 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1628 * processes read and write access respectively.
1630 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1632 struct fd f = fdget(fd);
1633 struct file_lock *lock;
1634 int can_sleep, unlock;
1641 can_sleep = !(cmd & LOCK_NB);
1643 unlock = (cmd == LOCK_UN);
1645 if (!unlock && !(cmd & LOCK_MAND) &&
1646 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1649 error = flock_make_lock(f.file, &lock, cmd);
1653 lock->fl_flags |= FL_SLEEP;
1655 error = security_file_lock(f.file, lock->fl_type);
1659 if (f.file->f_op && f.file->f_op->flock)
1660 error = f.file->f_op->flock(f.file,
1661 (can_sleep) ? F_SETLKW : F_SETLK,
1664 error = flock_lock_file_wait(f.file, lock);
1667 locks_free_lock(lock);
1676 * vfs_test_lock - test file byte range lock
1677 * @filp: The file to test lock for
1678 * @fl: The lock to test; also used to hold result
1680 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1681 * setting conf->fl_type to something other than F_UNLCK.
1683 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1685 if (filp->f_op && filp->f_op->lock)
1686 return filp->f_op->lock(filp, F_GETLK, fl);
1687 posix_test_lock(filp, fl);
1690 EXPORT_SYMBOL_GPL(vfs_test_lock);
1692 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1694 flock->l_pid = fl->fl_pid;
1695 #if BITS_PER_LONG == 32
1697 * Make sure we can represent the posix lock via
1698 * legacy 32bit flock.
1700 if (fl->fl_start > OFFT_OFFSET_MAX)
1702 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
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 #if BITS_PER_LONG == 32
1714 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1716 flock->l_pid = fl->fl_pid;
1717 flock->l_start = fl->fl_start;
1718 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1719 fl->fl_end - fl->fl_start + 1;
1720 flock->l_whence = 0;
1721 flock->l_type = fl->fl_type;
1725 /* Report the first existing lock that would conflict with l.
1726 * This implements the F_GETLK command of fcntl().
1728 int fcntl_getlk(struct file *filp, struct flock __user *l)
1730 struct file_lock file_lock;
1735 if (copy_from_user(&flock, l, sizeof(flock)))
1738 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1741 error = flock_to_posix_lock(filp, &file_lock, &flock);
1745 error = vfs_test_lock(filp, &file_lock);
1749 flock.l_type = file_lock.fl_type;
1750 if (file_lock.fl_type != F_UNLCK) {
1751 error = posix_lock_to_flock(&flock, &file_lock);
1756 if (!copy_to_user(l, &flock, sizeof(flock)))
1763 * vfs_lock_file - file byte range lock
1764 * @filp: The file to apply the lock to
1765 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1766 * @fl: The lock to be applied
1767 * @conf: Place to return a copy of the conflicting lock, if found.
1769 * A caller that doesn't care about the conflicting lock may pass NULL
1770 * as the final argument.
1772 * If the filesystem defines a private ->lock() method, then @conf will
1773 * be left unchanged; so a caller that cares should initialize it to
1774 * some acceptable default.
1776 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1777 * locks, the ->lock() interface may return asynchronously, before the lock has
1778 * been granted or denied by the underlying filesystem, if (and only if)
1779 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1780 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1781 * the request is for a blocking lock. When ->lock() does return asynchronously,
1782 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1783 * request completes.
1784 * If the request is for non-blocking lock the file system should return
1785 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1786 * with the result. If the request timed out the callback routine will return a
1787 * nonzero return code and the file system should release the lock. The file
1788 * system is also responsible to keep a corresponding posix lock when it
1789 * grants a lock so the VFS can find out which locks are locally held and do
1790 * the correct lock cleanup when required.
1791 * The underlying filesystem must not drop the kernel lock or call
1792 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1795 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1797 if (filp->f_op && filp->f_op->lock)
1798 return filp->f_op->lock(filp, cmd, fl);
1800 return posix_lock_file(filp, fl, conf);
1802 EXPORT_SYMBOL_GPL(vfs_lock_file);
1804 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1805 struct file_lock *fl)
1809 error = security_file_lock(filp, fl->fl_type);
1814 error = vfs_lock_file(filp, cmd, fl, NULL);
1815 if (error != FILE_LOCK_DEFERRED)
1817 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1821 locks_delete_block(fl);
1828 /* Apply the lock described by l to an open file descriptor.
1829 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1831 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1832 struct flock __user *l)
1834 struct file_lock *file_lock = locks_alloc_lock();
1836 struct inode *inode;
1840 if (file_lock == NULL)
1844 * This might block, so we do it before checking the inode.
1847 if (copy_from_user(&flock, l, sizeof(flock)))
1850 inode = file_inode(filp);
1852 /* Don't allow mandatory locks on files that may be memory mapped
1855 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1861 error = flock_to_posix_lock(filp, file_lock, &flock);
1864 if (cmd == F_SETLKW) {
1865 file_lock->fl_flags |= FL_SLEEP;
1869 switch (flock.l_type) {
1871 if (!(filp->f_mode & FMODE_READ))
1875 if (!(filp->f_mode & FMODE_WRITE))
1885 error = do_lock_file_wait(filp, cmd, file_lock);
1888 * Attempt to detect a close/fcntl race and recover by
1889 * releasing the lock that was just acquired.
1892 * we need that spin_lock here - it prevents reordering between
1893 * update of inode->i_flock and check for it done in close().
1894 * rcu_read_lock() wouldn't do.
1896 spin_lock(¤t->files->file_lock);
1898 spin_unlock(¤t->files->file_lock);
1899 if (!error && f != filp && flock.l_type != F_UNLCK) {
1900 flock.l_type = F_UNLCK;
1905 locks_free_lock(file_lock);
1909 #if BITS_PER_LONG == 32
1910 /* Report the first existing lock that would conflict with l.
1911 * This implements the F_GETLK command of fcntl().
1913 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1915 struct file_lock file_lock;
1916 struct flock64 flock;
1920 if (copy_from_user(&flock, l, sizeof(flock)))
1923 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1926 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1930 error = vfs_test_lock(filp, &file_lock);
1934 flock.l_type = file_lock.fl_type;
1935 if (file_lock.fl_type != F_UNLCK)
1936 posix_lock_to_flock64(&flock, &file_lock);
1939 if (!copy_to_user(l, &flock, sizeof(flock)))
1946 /* Apply the lock described by l to an open file descriptor.
1947 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1949 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1950 struct flock64 __user *l)
1952 struct file_lock *file_lock = locks_alloc_lock();
1953 struct flock64 flock;
1954 struct inode *inode;
1958 if (file_lock == NULL)
1962 * This might block, so we do it before checking the inode.
1965 if (copy_from_user(&flock, l, sizeof(flock)))
1968 inode = file_inode(filp);
1970 /* Don't allow mandatory locks on files that may be memory mapped
1973 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1979 error = flock64_to_posix_lock(filp, file_lock, &flock);
1982 if (cmd == F_SETLKW64) {
1983 file_lock->fl_flags |= FL_SLEEP;
1987 switch (flock.l_type) {
1989 if (!(filp->f_mode & FMODE_READ))
1993 if (!(filp->f_mode & FMODE_WRITE))
2003 error = do_lock_file_wait(filp, cmd, file_lock);
2006 * Attempt to detect a close/fcntl race and recover by
2007 * releasing the lock that was just acquired.
2009 spin_lock(¤t->files->file_lock);
2011 spin_unlock(¤t->files->file_lock);
2012 if (!error && f != filp && flock.l_type != F_UNLCK) {
2013 flock.l_type = F_UNLCK;
2018 locks_free_lock(file_lock);
2021 #endif /* BITS_PER_LONG == 32 */
2024 * This function is called when the file is being removed
2025 * from the task's fd array. POSIX locks belonging to this task
2026 * are deleted at this time.
2028 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2030 struct file_lock lock;
2033 * If there are no locks held on this file, we don't need to call
2034 * posix_lock_file(). Another process could be setting a lock on this
2035 * file at the same time, but we wouldn't remove that lock anyway.
2037 if (!file_inode(filp)->i_flock)
2040 lock.fl_type = F_UNLCK;
2041 lock.fl_flags = FL_POSIX | FL_CLOSE;
2043 lock.fl_end = OFFSET_MAX;
2044 lock.fl_owner = owner;
2045 lock.fl_pid = current->tgid;
2046 lock.fl_file = filp;
2048 lock.fl_lmops = NULL;
2050 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2052 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2053 lock.fl_ops->fl_release_private(&lock);
2056 EXPORT_SYMBOL(locks_remove_posix);
2059 * This function is called on the last close of an open file.
2061 void locks_remove_flock(struct file *filp)
2063 struct inode * inode = file_inode(filp);
2064 struct file_lock *fl;
2065 struct file_lock **before;
2067 if (!inode->i_flock)
2070 if (filp->f_op && filp->f_op->flock) {
2071 struct file_lock fl = {
2072 .fl_pid = current->tgid,
2074 .fl_flags = FL_FLOCK,
2076 .fl_end = OFFSET_MAX,
2078 filp->f_op->flock(filp, F_SETLKW, &fl);
2079 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2080 fl.fl_ops->fl_release_private(&fl);
2084 before = &inode->i_flock;
2086 while ((fl = *before) != NULL) {
2087 if (fl->fl_file == filp) {
2089 locks_delete_lock(before);
2093 lease_modify(before, F_UNLCK);
2099 before = &fl->fl_next;
2105 * posix_unblock_lock - stop waiting for a file lock
2106 * @waiter: the lock which was waiting
2108 * lockd needs to block waiting for locks.
2111 posix_unblock_lock(struct file_lock *waiter)
2116 if (waiter->fl_next)
2117 __locks_delete_block(waiter);
2123 EXPORT_SYMBOL(posix_unblock_lock);
2126 * vfs_cancel_lock - file byte range unblock lock
2127 * @filp: The file to apply the unblock to
2128 * @fl: The lock to be unblocked
2130 * Used by lock managers to cancel blocked requests
2132 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2134 if (filp->f_op && filp->f_op->lock)
2135 return filp->f_op->lock(filp, F_CANCELLK, fl);
2139 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2141 #ifdef CONFIG_PROC_FS
2142 #include <linux/proc_fs.h>
2143 #include <linux/seq_file.h>
2145 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2146 loff_t id, char *pfx)
2148 struct inode *inode = NULL;
2149 unsigned int fl_pid;
2152 fl_pid = pid_vnr(fl->fl_nspid);
2154 fl_pid = fl->fl_pid;
2156 if (fl->fl_file != NULL)
2157 inode = file_inode(fl->fl_file);
2159 seq_printf(f, "%lld:%s ", id, pfx);
2161 seq_printf(f, "%6s %s ",
2162 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2163 (inode == NULL) ? "*NOINODE*" :
2164 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2165 } else if (IS_FLOCK(fl)) {
2166 if (fl->fl_type & LOCK_MAND) {
2167 seq_printf(f, "FLOCK MSNFS ");
2169 seq_printf(f, "FLOCK ADVISORY ");
2171 } else if (IS_LEASE(fl)) {
2172 seq_printf(f, "LEASE ");
2173 if (lease_breaking(fl))
2174 seq_printf(f, "BREAKING ");
2175 else if (fl->fl_file)
2176 seq_printf(f, "ACTIVE ");
2178 seq_printf(f, "BREAKER ");
2180 seq_printf(f, "UNKNOWN UNKNOWN ");
2182 if (fl->fl_type & LOCK_MAND) {
2183 seq_printf(f, "%s ",
2184 (fl->fl_type & LOCK_READ)
2185 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2186 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2188 seq_printf(f, "%s ",
2189 (lease_breaking(fl))
2190 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2191 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2194 #ifdef WE_CAN_BREAK_LSLK_NOW
2195 seq_printf(f, "%d %s:%ld ", fl_pid,
2196 inode->i_sb->s_id, inode->i_ino);
2198 /* userspace relies on this representation of dev_t ;-( */
2199 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2200 MAJOR(inode->i_sb->s_dev),
2201 MINOR(inode->i_sb->s_dev), inode->i_ino);
2204 seq_printf(f, "%d <none>:0 ", fl_pid);
2207 if (fl->fl_end == OFFSET_MAX)
2208 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2210 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2212 seq_printf(f, "0 EOF\n");
2216 static int locks_show(struct seq_file *f, void *v)
2218 struct file_lock *fl, *bfl;
2220 fl = list_entry(v, struct file_lock, fl_link);
2222 lock_get_status(f, fl, *((loff_t *)f->private), "");
2224 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2225 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2230 static void *locks_start(struct seq_file *f, loff_t *pos)
2232 loff_t *p = f->private;
2236 return seq_list_start(&file_lock_list, *pos);
2239 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2241 loff_t *p = f->private;
2243 return seq_list_next(v, &file_lock_list, pos);
2246 static void locks_stop(struct seq_file *f, void *v)
2251 static const struct seq_operations locks_seq_operations = {
2252 .start = locks_start,
2258 static int locks_open(struct inode *inode, struct file *filp)
2260 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2263 static const struct file_operations proc_locks_operations = {
2266 .llseek = seq_lseek,
2267 .release = seq_release_private,
2270 static int __init proc_locks_init(void)
2272 proc_create("locks", 0, NULL, &proc_locks_operations);
2275 module_init(proc_locks_init);
2279 * lock_may_read - checks that the region is free of locks
2280 * @inode: the inode that is being read
2281 * @start: the first byte to read
2282 * @len: the number of bytes to read
2284 * Emulates Windows locking requirements. Whole-file
2285 * mandatory locks (share modes) can prohibit a read and
2286 * byte-range POSIX locks can prohibit a read if they overlap.
2288 * N.B. this function is only ever called
2289 * from knfsd and ownership of locks is never checked.
2291 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2293 struct file_lock *fl;
2296 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2298 if (fl->fl_type == F_RDLCK)
2300 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2302 } else if (IS_FLOCK(fl)) {
2303 if (!(fl->fl_type & LOCK_MAND))
2305 if (fl->fl_type & LOCK_READ)
2316 EXPORT_SYMBOL(lock_may_read);
2319 * lock_may_write - checks that the region is free of locks
2320 * @inode: the inode that is being written
2321 * @start: the first byte to write
2322 * @len: the number of bytes to write
2324 * Emulates Windows locking requirements. Whole-file
2325 * mandatory locks (share modes) can prohibit a write and
2326 * byte-range POSIX locks can prohibit a write if they overlap.
2328 * N.B. this function is only ever called
2329 * from knfsd and ownership of locks is never checked.
2331 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2333 struct file_lock *fl;
2336 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2338 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2340 } else if (IS_FLOCK(fl)) {
2341 if (!(fl->fl_type & LOCK_MAND))
2343 if (fl->fl_type & LOCK_WRITE)
2354 EXPORT_SYMBOL(lock_may_write);
2356 static int __init filelock_init(void)
2358 filelock_cache = kmem_cache_create("file_lock_cache",
2359 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2364 core_initcall(filelock_init);