4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
129 #include <asm/semaphore.h>
130 #include <asm/uaccess.h>
132 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136 int leases_enable = 1;
137 int lease_break_time = 45;
139 #define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
145 static kmem_cache_t *filelock_cache __read_mostly;
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
150 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
153 static void locks_release_private(struct file_lock *fl)
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
175 locks_release_private(fl);
176 kmem_cache_free(filelock_cache, fl);
179 void locks_init_lock(struct file_lock *fl)
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
185 fl->fl_fasync = NULL;
191 fl->fl_start = fl->fl_end = 0;
196 EXPORT_SYMBOL(locks_init_lock);
199 * Initialises the fields of the file lock which are invariant for
202 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
204 struct file_lock *lock = (struct file_lock *) foo;
206 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
207 SLAB_CTOR_CONSTRUCTOR)
210 locks_init_lock(lock);
213 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
216 if (fl->fl_ops->fl_copy_lock)
217 fl->fl_ops->fl_copy_lock(new, fl);
218 new->fl_ops = fl->fl_ops;
221 if (fl->fl_lmops->fl_copy_lock)
222 fl->fl_lmops->fl_copy_lock(new, fl);
223 new->fl_lmops = fl->fl_lmops;
228 * Initialize a new lock from an existing file_lock structure.
230 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
232 new->fl_owner = fl->fl_owner;
233 new->fl_pid = fl->fl_pid;
235 new->fl_flags = fl->fl_flags;
236 new->fl_type = fl->fl_type;
237 new->fl_start = fl->fl_start;
238 new->fl_end = fl->fl_end;
240 new->fl_lmops = NULL;
243 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
245 locks_release_private(new);
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
252 locks_copy_private(new, fl);
255 EXPORT_SYMBOL(locks_copy_lock);
257 static inline int flock_translate_cmd(int cmd) {
259 return cmd & (LOCK_MAND | LOCK_RW);
271 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
272 static int flock_make_lock(struct file *filp, struct file_lock **lock,
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
280 fl = locks_alloc_lock();
285 fl->fl_pid = current->tgid;
286 fl->fl_flags = FL_FLOCK;
288 fl->fl_end = OFFSET_MAX;
294 static int assign_type(struct file_lock *fl, int type)
308 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
311 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
316 switch (l->l_whence) {
324 start = i_size_read(filp->f_dentry->d_inode);
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
335 fl->fl_end = OFFSET_MAX;
337 end = start + l->l_len - 1;
339 } else if (l->l_len < 0) {
346 fl->fl_start = start; /* we record the absolute position */
347 if (fl->fl_end < fl->fl_start)
350 fl->fl_owner = current->files;
351 fl->fl_pid = current->tgid;
353 fl->fl_flags = FL_POSIX;
357 return assign_type(fl, l->l_type);
360 #if BITS_PER_LONG == 32
361 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
366 switch (l->l_whence) {
374 start = i_size_read(filp->f_dentry->d_inode);
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_end = start + l->l_len - 1;
386 } else if (l->l_len < 0) {
387 fl->fl_end = start - 1;
392 fl->fl_start = start; /* we record the absolute position */
393 if (fl->fl_end < fl->fl_start)
396 fl->fl_owner = current->files;
397 fl->fl_pid = current->tgid;
399 fl->fl_flags = FL_POSIX;
407 fl->fl_type = l->l_type;
417 /* default lease lock manager operations */
418 static void lease_break_callback(struct file_lock *fl)
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
423 static void lease_release_private_callback(struct file_lock *fl)
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
432 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
434 return fl->fl_file == try->fl_file;
437 static struct lock_manager_operations lease_manager_ops = {
438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
445 * Initialize a lease, use the default lock manager operations
447 static int lease_init(struct file *filp, int type, struct file_lock *fl)
449 if (assign_type(fl, type) != 0)
452 fl->fl_owner = current->files;
453 fl->fl_pid = current->tgid;
456 fl->fl_flags = FL_LEASE;
458 fl->fl_end = OFFSET_MAX;
460 fl->fl_lmops = &lease_manager_ops;
464 /* Allocate a file_lock initialised to this type of lease */
465 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
467 struct file_lock *fl = locks_alloc_lock();
473 error = lease_init(filp, type, fl);
483 /* Check if two locks overlap each other.
485 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
487 return ((fl1->fl_end >= fl2->fl_start) &&
488 (fl2->fl_end >= fl1->fl_start));
492 * Check whether two locks have the same owner.
494 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
496 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
497 return fl2->fl_lmops == fl1->fl_lmops &&
498 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
499 return fl1->fl_owner == fl2->fl_owner;
502 /* Remove waiter from blocker's block list.
503 * When blocker ends up pointing to itself then the list is empty.
505 static void __locks_delete_block(struct file_lock *waiter)
507 list_del_init(&waiter->fl_block);
508 list_del_init(&waiter->fl_link);
509 waiter->fl_next = NULL;
514 static void locks_delete_block(struct file_lock *waiter)
517 __locks_delete_block(waiter);
521 /* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
526 static void locks_insert_block(struct file_lock *blocker,
527 struct file_lock *waiter)
529 BUG_ON(!list_empty(&waiter->fl_block));
530 list_add_tail(&waiter->fl_block, &blocker->fl_block);
531 waiter->fl_next = blocker;
532 if (IS_POSIX(blocker))
533 list_add(&waiter->fl_link, &blocked_list);
536 /* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
540 static void locks_wake_up_blocks(struct file_lock *blocker)
542 while (!list_empty(&blocker->fl_block)) {
543 struct file_lock *waiter = list_entry(blocker->fl_block.next,
544 struct file_lock, fl_block);
545 __locks_delete_block(waiter);
546 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
547 waiter->fl_lmops->fl_notify(waiter);
549 wake_up(&waiter->fl_wait);
553 /* Insert file lock fl into an inode's lock list at the position indicated
554 * by pos. At the same time add the lock to the global file lock list.
556 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
558 list_add(&fl->fl_link, &file_lock_list);
560 /* insert into file's list */
564 if (fl->fl_ops && fl->fl_ops->fl_insert)
565 fl->fl_ops->fl_insert(fl);
569 * Delete a lock and then free it.
570 * Wake up processes that are blocked waiting for this lock,
571 * notify the FS that the lock has been cleared and
572 * finally free the lock.
574 static void locks_delete_lock(struct file_lock **thisfl_p)
576 struct file_lock *fl = *thisfl_p;
578 *thisfl_p = fl->fl_next;
580 list_del_init(&fl->fl_link);
582 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
583 if (fl->fl_fasync != NULL) {
584 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
585 fl->fl_fasync = NULL;
588 if (fl->fl_ops && fl->fl_ops->fl_remove)
589 fl->fl_ops->fl_remove(fl);
591 locks_wake_up_blocks(fl);
595 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
596 * checks for shared/exclusive status of overlapping locks.
598 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
600 if (sys_fl->fl_type == F_WRLCK)
602 if (caller_fl->fl_type == F_WRLCK)
607 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
608 * checking before calling the locks_conflict().
610 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
612 /* POSIX locks owned by the same process do not conflict with
615 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
618 /* Check whether they overlap */
619 if (!locks_overlap(caller_fl, sys_fl))
622 return (locks_conflict(caller_fl, sys_fl));
625 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
626 * checking before calling the locks_conflict().
628 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
630 /* FLOCK locks referring to the same filp do not conflict with
633 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
635 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
638 return (locks_conflict(caller_fl, sys_fl));
641 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
644 DECLARE_WAITQUEUE(wait, current);
646 __set_current_state(TASK_INTERRUPTIBLE);
647 add_wait_queue(fl_wait, &wait);
651 result = schedule_timeout(timeout);
652 if (signal_pending(current))
653 result = -ERESTARTSYS;
654 remove_wait_queue(fl_wait, &wait);
655 __set_current_state(TASK_RUNNING);
659 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
662 locks_insert_block(blocker, waiter);
663 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
664 __locks_delete_block(waiter);
669 posix_test_lock(struct file *filp, struct file_lock *fl,
670 struct file_lock *conflock)
672 struct file_lock *cfl;
675 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
678 if (posix_locks_conflict(cfl, fl))
682 __locks_copy_lock(conflock, cfl);
690 EXPORT_SYMBOL(posix_test_lock);
692 /* This function tests for deadlock condition before putting a process to
693 * sleep. The detection scheme is no longer recursive. Recursive was neat,
694 * but dangerous - we risked stack corruption if the lock data was bad, or
695 * if the recursion was too deep for any other reason.
697 * We rely on the fact that a task can only be on one lock's wait queue
698 * at a time. When we find blocked_task on a wait queue we can re-search
699 * with blocked_task equal to that queue's owner, until either blocked_task
700 * isn't found, or blocked_task is found on a queue owned by my_task.
702 * Note: the above assumption may not be true when handling lock requests
703 * from a broken NFS client. But broken NFS clients have a lot more to
704 * worry about than proper deadlock detection anyway... --okir
706 int posix_locks_deadlock(struct file_lock *caller_fl,
707 struct file_lock *block_fl)
709 struct list_head *tmp;
712 if (posix_same_owner(caller_fl, block_fl))
714 list_for_each(tmp, &blocked_list) {
715 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
716 if (posix_same_owner(fl, block_fl)) {
725 EXPORT_SYMBOL(posix_locks_deadlock);
727 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
728 * at the head of the list, but that's secret knowledge known only to
729 * flock_lock_file and posix_lock_file.
731 static int flock_lock_file(struct file *filp, struct file_lock *request)
733 struct file_lock *new_fl = NULL;
734 struct file_lock **before;
735 struct inode * inode = filp->f_dentry->d_inode;
740 for_each_lock(inode, before) {
741 struct file_lock *fl = *before;
746 if (filp != fl->fl_file)
748 if (request->fl_type == fl->fl_type)
751 locks_delete_lock(before);
755 if (request->fl_type == F_UNLCK)
758 new_fl = locks_alloc_lock();
762 * If a higher-priority process was blocked on the old file lock,
763 * give it the opportunity to lock the file.
768 for_each_lock(inode, before) {
769 struct file_lock *fl = *before;
774 if (!flock_locks_conflict(request, fl))
777 if (request->fl_flags & FL_SLEEP)
778 locks_insert_block(fl, request);
781 locks_copy_lock(new_fl, request);
782 locks_insert_lock(&inode->i_flock, new_fl);
788 locks_free_lock(new_fl);
792 static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
794 struct file_lock *fl;
795 struct file_lock *new_fl, *new_fl2;
796 struct file_lock *left = NULL;
797 struct file_lock *right = NULL;
798 struct file_lock **before;
799 int error, added = 0;
802 * We may need two file_lock structures for this operation,
803 * so we get them in advance to avoid races.
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
809 if (request->fl_type != F_UNLCK) {
810 for_each_lock(inode, before) {
811 struct file_lock *fl = *before;
814 if (!posix_locks_conflict(request, fl))
817 locks_copy_lock(conflock, fl);
819 if (!(request->fl_flags & FL_SLEEP))
822 if (posix_locks_deadlock(request, fl))
825 locks_insert_block(fl, request);
830 /* If we're just looking for a conflict, we're done. */
832 if (request->fl_flags & FL_ACCESS)
835 error = -ENOLCK; /* "no luck" */
836 if (!(new_fl && new_fl2))
840 * We've allocated the new locks in advance, so there are no
841 * errors possible (and no blocking operations) from here on.
843 * Find the first old lock with the same owner as the new lock.
846 before = &inode->i_flock;
848 /* First skip locks owned by other processes. */
849 while ((fl = *before) && (!IS_POSIX(fl) ||
850 !posix_same_owner(request, fl))) {
851 before = &fl->fl_next;
854 /* Process locks with this owner. */
855 while ((fl = *before) && posix_same_owner(request, fl)) {
856 /* Detect adjacent or overlapping regions (if same lock type)
858 if (request->fl_type == fl->fl_type) {
859 /* In all comparisons of start vs end, use
860 * "start - 1" rather than "end + 1". If end
861 * is OFFSET_MAX, end + 1 will become negative.
863 if (fl->fl_end < request->fl_start - 1)
865 /* If the next lock in the list has entirely bigger
866 * addresses than the new one, insert the lock here.
868 if (fl->fl_start - 1 > request->fl_end)
871 /* If we come here, the new and old lock are of the
872 * same type and adjacent or overlapping. Make one
873 * lock yielding from the lower start address of both
874 * locks to the higher end address.
876 if (fl->fl_start > request->fl_start)
877 fl->fl_start = request->fl_start;
879 request->fl_start = fl->fl_start;
880 if (fl->fl_end < request->fl_end)
881 fl->fl_end = request->fl_end;
883 request->fl_end = fl->fl_end;
885 locks_delete_lock(before);
892 /* Processing for different lock types is a bit
895 if (fl->fl_end < request->fl_start)
897 if (fl->fl_start > request->fl_end)
899 if (request->fl_type == F_UNLCK)
901 if (fl->fl_start < request->fl_start)
903 /* If the next lock in the list has a higher end
904 * address than the new one, insert the new one here.
906 if (fl->fl_end > request->fl_end) {
910 if (fl->fl_start >= request->fl_start) {
911 /* The new lock completely replaces an old
912 * one (This may happen several times).
915 locks_delete_lock(before);
918 /* Replace the old lock with the new one.
919 * Wake up anybody waiting for the old one,
920 * as the change in lock type might satisfy
923 locks_wake_up_blocks(fl);
924 fl->fl_start = request->fl_start;
925 fl->fl_end = request->fl_end;
926 fl->fl_type = request->fl_type;
927 locks_release_private(fl);
928 locks_copy_private(fl, request);
933 /* Go on to next lock.
936 before = &fl->fl_next;
941 if (request->fl_type == F_UNLCK)
943 locks_copy_lock(new_fl, request);
944 locks_insert_lock(before, new_fl);
949 /* The new lock breaks the old one in two pieces,
950 * so we have to use the second new lock.
954 locks_copy_lock(left, right);
955 locks_insert_lock(before, left);
957 right->fl_start = request->fl_end + 1;
958 locks_wake_up_blocks(right);
961 left->fl_end = request->fl_start - 1;
962 locks_wake_up_blocks(left);
967 * Free any unused locks.
970 locks_free_lock(new_fl);
972 locks_free_lock(new_fl2);
977 * posix_lock_file - Apply a POSIX-style lock to a file
978 * @filp: The file to apply the lock to
979 * @fl: The lock to be applied
981 * Add a POSIX style lock to a file.
982 * We merge adjacent & overlapping locks whenever possible.
983 * POSIX locks are sorted by owner task, then by starting address
985 int posix_lock_file(struct file *filp, struct file_lock *fl)
987 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
989 EXPORT_SYMBOL(posix_lock_file);
992 * posix_lock_file_conf - Apply a POSIX-style lock to a file
993 * @filp: The file to apply the lock to
994 * @fl: The lock to be applied
995 * @conflock: Place to return a copy of the conflicting lock, if found.
997 * Except for the conflock parameter, acts just like posix_lock_file.
999 int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1000 struct file_lock *conflock)
1002 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1004 EXPORT_SYMBOL(posix_lock_file_conf);
1007 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1008 * @filp: The file to apply the lock to
1009 * @fl: The lock to be applied
1011 * Add a POSIX style lock to a file.
1012 * We merge adjacent & overlapping locks whenever possible.
1013 * POSIX locks are sorted by owner task, then by starting address
1015 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1020 error = posix_lock_file(filp, fl);
1021 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1023 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1027 locks_delete_block(fl);
1032 EXPORT_SYMBOL(posix_lock_file_wait);
1035 * locks_mandatory_locked - Check for an active lock
1036 * @inode: the file to check
1038 * Searches the inode's list of locks to find any POSIX locks which conflict.
1039 * This function is called from locks_verify_locked() only.
1041 int locks_mandatory_locked(struct inode *inode)
1043 fl_owner_t owner = current->files;
1044 struct file_lock *fl;
1047 * Search the lock list for this inode for any POSIX locks.
1050 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1053 if (fl->fl_owner != owner)
1057 return fl ? -EAGAIN : 0;
1061 * locks_mandatory_area - Check for a conflicting lock
1062 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1064 * @inode: the file to check
1065 * @filp: how the file was opened (if it was)
1066 * @offset: start of area to check
1067 * @count: length of area to check
1069 * Searches the inode's list of locks to find any POSIX locks which conflict.
1070 * This function is called from rw_verify_area() and
1071 * locks_verify_truncate().
1073 int locks_mandatory_area(int read_write, struct inode *inode,
1074 struct file *filp, loff_t offset,
1077 struct file_lock fl;
1080 locks_init_lock(&fl);
1081 fl.fl_owner = current->files;
1082 fl.fl_pid = current->tgid;
1084 fl.fl_flags = FL_POSIX | FL_ACCESS;
1085 if (filp && !(filp->f_flags & O_NONBLOCK))
1086 fl.fl_flags |= FL_SLEEP;
1087 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1088 fl.fl_start = offset;
1089 fl.fl_end = offset + count - 1;
1092 error = __posix_lock_file_conf(inode, &fl, NULL);
1093 if (error != -EAGAIN)
1095 if (!(fl.fl_flags & FL_SLEEP))
1097 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1100 * If we've been sleeping someone might have
1101 * changed the permissions behind our back.
1103 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1107 locks_delete_block(&fl);
1114 EXPORT_SYMBOL(locks_mandatory_area);
1116 /* We already had a lease on this file; just change its type */
1117 int lease_modify(struct file_lock **before, int arg)
1119 struct file_lock *fl = *before;
1120 int error = assign_type(fl, arg);
1124 locks_wake_up_blocks(fl);
1126 locks_delete_lock(before);
1130 EXPORT_SYMBOL(lease_modify);
1132 static void time_out_leases(struct inode *inode)
1134 struct file_lock **before;
1135 struct file_lock *fl;
1137 before = &inode->i_flock;
1138 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1139 if ((fl->fl_break_time == 0)
1140 || time_before(jiffies, fl->fl_break_time)) {
1141 before = &fl->fl_next;
1144 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1145 if (fl == *before) /* lease_modify may have freed fl */
1146 before = &fl->fl_next;
1151 * __break_lease - revoke all outstanding leases on file
1152 * @inode: the inode of the file to return
1153 * @mode: the open mode (read or write)
1155 * break_lease (inlined for speed) has checked there already
1156 * is a lease on this file. Leases are broken on a call to open()
1157 * or truncate(). This function can sleep unless you
1158 * specified %O_NONBLOCK to your open().
1160 int __break_lease(struct inode *inode, unsigned int mode)
1162 int error = 0, future;
1163 struct file_lock *new_fl, *flock;
1164 struct file_lock *fl;
1166 unsigned long break_time;
1167 int i_have_this_lease = 0;
1169 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1174 time_out_leases(inode);
1176 flock = inode->i_flock;
1177 if ((flock == NULL) || !IS_LEASE(flock))
1180 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1181 if (fl->fl_owner == current->files)
1182 i_have_this_lease = 1;
1184 if (mode & FMODE_WRITE) {
1185 /* If we want write access, we have to revoke any lease. */
1186 future = F_UNLCK | F_INPROGRESS;
1187 } else if (flock->fl_type & F_INPROGRESS) {
1188 /* If the lease is already being broken, we just leave it */
1189 future = flock->fl_type;
1190 } else if (flock->fl_type & F_WRLCK) {
1191 /* Downgrade the exclusive lease to a read-only lease. */
1192 future = F_RDLCK | F_INPROGRESS;
1194 /* the existing lease was read-only, so we can read too. */
1198 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1204 if (lease_break_time > 0) {
1205 break_time = jiffies + lease_break_time * HZ;
1206 if (break_time == 0)
1207 break_time++; /* so that 0 means no break time */
1210 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1211 if (fl->fl_type != future) {
1212 fl->fl_type = future;
1213 fl->fl_break_time = break_time;
1214 /* lease must have lmops break callback */
1215 fl->fl_lmops->fl_break(fl);
1219 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1220 error = -EWOULDBLOCK;
1225 break_time = flock->fl_break_time;
1226 if (break_time != 0) {
1227 break_time -= jiffies;
1228 if (break_time == 0)
1231 error = locks_block_on_timeout(flock, new_fl, break_time);
1234 time_out_leases(inode);
1235 /* Wait for the next lease that has not been broken yet */
1236 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1237 flock = flock->fl_next) {
1238 if (flock->fl_type & F_INPROGRESS)
1247 locks_free_lock(new_fl);
1251 EXPORT_SYMBOL(__break_lease);
1256 * @time: pointer to a timespec which will contain the last modified time
1258 * This is to force NFS clients to flush their caches for files with
1259 * exclusive leases. The justification is that if someone has an
1260 * exclusive lease, then they could be modifiying it.
1262 void lease_get_mtime(struct inode *inode, struct timespec *time)
1264 struct file_lock *flock = inode->i_flock;
1265 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1266 *time = current_fs_time(inode->i_sb);
1268 *time = inode->i_mtime;
1271 EXPORT_SYMBOL(lease_get_mtime);
1274 * fcntl_getlease - Enquire what lease is currently active
1277 * The value returned by this function will be one of
1278 * (if no lease break is pending):
1280 * %F_RDLCK to indicate a shared lease is held.
1282 * %F_WRLCK to indicate an exclusive lease is held.
1284 * %F_UNLCK to indicate no lease is held.
1286 * (if a lease break is pending):
1288 * %F_RDLCK to indicate an exclusive lease needs to be
1289 * changed to a shared lease (or removed).
1291 * %F_UNLCK to indicate the lease needs to be removed.
1293 * XXX: sfr & willy disagree over whether F_INPROGRESS
1294 * should be returned to userspace.
1296 int fcntl_getlease(struct file *filp)
1298 struct file_lock *fl;
1302 time_out_leases(filp->f_dentry->d_inode);
1303 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1305 if (fl->fl_file == filp) {
1306 type = fl->fl_type & ~F_INPROGRESS;
1315 * __setlease - sets a lease on an open file
1316 * @filp: file pointer
1317 * @arg: type of lease to obtain
1318 * @flp: input - file_lock to use, output - file_lock inserted
1320 * The (input) flp->fl_lmops->fl_break function is required
1323 * Called with kernel lock held.
1325 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1327 struct file_lock *fl, **before, **my_before = NULL, *lease;
1328 struct dentry *dentry = filp->f_dentry;
1329 struct inode *inode = dentry->d_inode;
1330 int error, rdlease_count = 0, wrlease_count = 0;
1332 time_out_leases(inode);
1335 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1341 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1343 if ((arg == F_WRLCK)
1344 && ((atomic_read(&dentry->d_count) > 1)
1345 || (atomic_read(&inode->i_count) > 1)))
1349 * At this point, we know that if there is an exclusive
1350 * lease on this file, then we hold it on this filp
1351 * (otherwise our open of this file would have blocked).
1352 * And if we are trying to acquire an exclusive lease,
1353 * then the file is not open by anyone (including us)
1354 * except for this filp.
1356 for (before = &inode->i_flock;
1357 ((fl = *before) != NULL) && IS_LEASE(fl);
1358 before = &fl->fl_next) {
1359 if (lease->fl_lmops->fl_mylease(fl, lease))
1361 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1363 * Someone is in the process of opening this
1364 * file for writing so we may not take an
1365 * exclusive lease on it.
1372 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1373 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1376 if (my_before != NULL) {
1378 error = lease->fl_lmops->fl_change(my_before, arg);
1390 error = lease_alloc(filp, arg, &fl);
1394 locks_copy_lock(fl, lease);
1396 locks_insert_lock(before, fl);
1404 * setlease - sets a lease on an open file
1405 * @filp: file pointer
1406 * @arg: type of lease to obtain
1407 * @lease: file_lock to use
1409 * Call this to establish a lease on the file.
1410 * The fl_lmops fl_break function is required by break_lease
1413 int setlease(struct file *filp, long arg, struct file_lock **lease)
1415 struct dentry *dentry = filp->f_dentry;
1416 struct inode *inode = dentry->d_inode;
1419 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1421 if (!S_ISREG(inode->i_mode))
1423 error = security_file_lock(filp, arg);
1428 error = __setlease(filp, arg, lease);
1434 EXPORT_SYMBOL(setlease);
1437 * fcntl_setlease - sets a lease on an open file
1438 * @fd: open file descriptor
1439 * @filp: file pointer
1440 * @arg: type of lease to obtain
1442 * Call this fcntl to establish a lease on the file.
1443 * Note that you also need to call %F_SETSIG to
1444 * receive a signal when the lease is broken.
1446 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1448 struct file_lock fl, *flp = &fl;
1449 struct dentry *dentry = filp->f_dentry;
1450 struct inode *inode = dentry->d_inode;
1453 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1455 if (!S_ISREG(inode->i_mode))
1457 error = security_file_lock(filp, arg);
1461 locks_init_lock(&fl);
1462 error = lease_init(filp, arg, &fl);
1468 error = __setlease(filp, arg, &flp);
1469 if (error || arg == F_UNLCK)
1472 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1474 /* remove lease just inserted by __setlease */
1475 flp->fl_type = F_UNLCK | F_INPROGRESS;
1476 flp->fl_break_time = jiffies- 10;
1477 time_out_leases(inode);
1481 error = f_setown(filp, current->pid, 0);
1488 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1489 * @filp: The file to apply the lock to
1490 * @fl: The lock to be applied
1492 * Add a FLOCK style lock to a file.
1494 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1499 error = flock_lock_file(filp, fl);
1500 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1502 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1506 locks_delete_block(fl);
1512 EXPORT_SYMBOL(flock_lock_file_wait);
1515 * sys_flock: - flock() system call.
1516 * @fd: the file descriptor to lock.
1517 * @cmd: the type of lock to apply.
1519 * Apply a %FL_FLOCK style lock to an open file descriptor.
1520 * The @cmd can be one of
1522 * %LOCK_SH -- a shared lock.
1524 * %LOCK_EX -- an exclusive lock.
1526 * %LOCK_UN -- remove an existing lock.
1528 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1530 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1531 * processes read and write access respectively.
1533 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1536 struct file_lock *lock;
1537 int can_sleep, unlock;
1545 can_sleep = !(cmd & LOCK_NB);
1547 unlock = (cmd == LOCK_UN);
1549 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1552 error = flock_make_lock(filp, &lock, cmd);
1556 lock->fl_flags |= FL_SLEEP;
1558 error = security_file_lock(filp, cmd);
1562 if (filp->f_op && filp->f_op->flock)
1563 error = filp->f_op->flock(filp,
1564 (can_sleep) ? F_SETLKW : F_SETLK,
1567 error = flock_lock_file_wait(filp, lock);
1570 locks_free_lock(lock);
1578 /* Report the first existing lock that would conflict with l.
1579 * This implements the F_GETLK command of fcntl().
1581 int fcntl_getlk(struct file *filp, struct flock __user *l)
1583 struct file_lock *fl, cfl, file_lock;
1588 if (copy_from_user(&flock, l, sizeof(flock)))
1591 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1594 error = flock_to_posix_lock(filp, &file_lock, &flock);
1598 if (filp->f_op && filp->f_op->lock) {
1599 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1600 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1601 file_lock.fl_ops->fl_release_private(&file_lock);
1605 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1607 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1610 flock.l_type = F_UNLCK;
1612 flock.l_pid = fl->fl_pid;
1613 #if BITS_PER_LONG == 32
1615 * Make sure we can represent the posix lock via
1616 * legacy 32bit flock.
1619 if (fl->fl_start > OFFT_OFFSET_MAX)
1621 if ((fl->fl_end != OFFSET_MAX)
1622 && (fl->fl_end > OFFT_OFFSET_MAX))
1625 flock.l_start = fl->fl_start;
1626 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1627 fl->fl_end - fl->fl_start + 1;
1629 flock.l_type = fl->fl_type;
1632 if (!copy_to_user(l, &flock, sizeof(flock)))
1638 /* Apply the lock described by l to an open file descriptor.
1639 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1641 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1642 struct flock __user *l)
1644 struct file_lock *file_lock = locks_alloc_lock();
1646 struct inode *inode;
1649 if (file_lock == NULL)
1653 * This might block, so we do it before checking the inode.
1656 if (copy_from_user(&flock, l, sizeof(flock)))
1659 inode = filp->f_dentry->d_inode;
1661 /* Don't allow mandatory locks on files that may be memory mapped
1664 if (IS_MANDLOCK(inode) &&
1665 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1666 mapping_writably_mapped(filp->f_mapping)) {
1672 error = flock_to_posix_lock(filp, file_lock, &flock);
1675 if (cmd == F_SETLKW) {
1676 file_lock->fl_flags |= FL_SLEEP;
1680 switch (flock.l_type) {
1682 if (!(filp->f_mode & FMODE_READ))
1686 if (!(filp->f_mode & FMODE_WRITE))
1696 error = security_file_lock(filp, file_lock->fl_type);
1700 if (filp->f_op && filp->f_op->lock != NULL)
1701 error = filp->f_op->lock(filp, cmd, file_lock);
1704 error = posix_lock_file(filp, file_lock);
1705 if ((error != -EAGAIN) || (cmd == F_SETLK))
1707 error = wait_event_interruptible(file_lock->fl_wait,
1708 !file_lock->fl_next);
1712 locks_delete_block(file_lock);
1718 * Attempt to detect a close/fcntl race and recover by
1719 * releasing the lock that was just acquired.
1721 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1722 flock.l_type = F_UNLCK;
1727 locks_free_lock(file_lock);
1731 #if BITS_PER_LONG == 32
1732 /* Report the first existing lock that would conflict with l.
1733 * This implements the F_GETLK command of fcntl().
1735 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1737 struct file_lock *fl, cfl, file_lock;
1738 struct flock64 flock;
1742 if (copy_from_user(&flock, l, sizeof(flock)))
1745 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1748 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1752 if (filp->f_op && filp->f_op->lock) {
1753 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1754 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1755 file_lock.fl_ops->fl_release_private(&file_lock);
1759 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1761 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1764 flock.l_type = F_UNLCK;
1766 flock.l_pid = fl->fl_pid;
1767 flock.l_start = fl->fl_start;
1768 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1769 fl->fl_end - fl->fl_start + 1;
1771 flock.l_type = fl->fl_type;
1774 if (!copy_to_user(l, &flock, sizeof(flock)))
1781 /* Apply the lock described by l to an open file descriptor.
1782 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1784 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1785 struct flock64 __user *l)
1787 struct file_lock *file_lock = locks_alloc_lock();
1788 struct flock64 flock;
1789 struct inode *inode;
1792 if (file_lock == NULL)
1796 * This might block, so we do it before checking the inode.
1799 if (copy_from_user(&flock, l, sizeof(flock)))
1802 inode = filp->f_dentry->d_inode;
1804 /* Don't allow mandatory locks on files that may be memory mapped
1807 if (IS_MANDLOCK(inode) &&
1808 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1809 mapping_writably_mapped(filp->f_mapping)) {
1815 error = flock64_to_posix_lock(filp, file_lock, &flock);
1818 if (cmd == F_SETLKW64) {
1819 file_lock->fl_flags |= FL_SLEEP;
1823 switch (flock.l_type) {
1825 if (!(filp->f_mode & FMODE_READ))
1829 if (!(filp->f_mode & FMODE_WRITE))
1839 error = security_file_lock(filp, file_lock->fl_type);
1843 if (filp->f_op && filp->f_op->lock != NULL)
1844 error = filp->f_op->lock(filp, cmd, file_lock);
1847 error = posix_lock_file(filp, file_lock);
1848 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1850 error = wait_event_interruptible(file_lock->fl_wait,
1851 !file_lock->fl_next);
1855 locks_delete_block(file_lock);
1861 * Attempt to detect a close/fcntl race and recover by
1862 * releasing the lock that was just acquired.
1864 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1865 flock.l_type = F_UNLCK;
1870 locks_free_lock(file_lock);
1873 #endif /* BITS_PER_LONG == 32 */
1876 * This function is called when the file is being removed
1877 * from the task's fd array. POSIX locks belonging to this task
1878 * are deleted at this time.
1880 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1882 struct file_lock lock, **before;
1885 * If there are no locks held on this file, we don't need to call
1886 * posix_lock_file(). Another process could be setting a lock on this
1887 * file at the same time, but we wouldn't remove that lock anyway.
1889 before = &filp->f_dentry->d_inode->i_flock;
1890 if (*before == NULL)
1893 lock.fl_type = F_UNLCK;
1894 lock.fl_flags = FL_POSIX;
1896 lock.fl_end = OFFSET_MAX;
1897 lock.fl_owner = owner;
1898 lock.fl_pid = current->tgid;
1899 lock.fl_file = filp;
1901 lock.fl_lmops = NULL;
1903 if (filp->f_op && filp->f_op->lock != NULL) {
1904 filp->f_op->lock(filp, F_SETLK, &lock);
1908 /* Can't use posix_lock_file here; we need to remove it no matter
1909 * which pid we have.
1912 while (*before != NULL) {
1913 struct file_lock *fl = *before;
1914 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1915 locks_delete_lock(before);
1918 before = &fl->fl_next;
1922 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1923 lock.fl_ops->fl_release_private(&lock);
1926 EXPORT_SYMBOL(locks_remove_posix);
1929 * This function is called on the last close of an open file.
1931 void locks_remove_flock(struct file *filp)
1933 struct inode * inode = filp->f_dentry->d_inode;
1934 struct file_lock *fl;
1935 struct file_lock **before;
1937 if (!inode->i_flock)
1940 if (filp->f_op && filp->f_op->flock) {
1941 struct file_lock fl = {
1942 .fl_pid = current->tgid,
1944 .fl_flags = FL_FLOCK,
1946 .fl_end = OFFSET_MAX,
1948 filp->f_op->flock(filp, F_SETLKW, &fl);
1949 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1950 fl.fl_ops->fl_release_private(&fl);
1954 before = &inode->i_flock;
1956 while ((fl = *before) != NULL) {
1957 if (fl->fl_file == filp) {
1959 locks_delete_lock(before);
1963 lease_modify(before, F_UNLCK);
1969 before = &fl->fl_next;
1975 * posix_unblock_lock - stop waiting for a file lock
1976 * @filp: how the file was opened
1977 * @waiter: the lock which was waiting
1979 * lockd needs to block waiting for locks.
1982 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1987 if (waiter->fl_next)
1988 __locks_delete_block(waiter);
1995 EXPORT_SYMBOL(posix_unblock_lock);
1997 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1999 struct inode *inode = NULL;
2001 if (fl->fl_file != NULL)
2002 inode = fl->fl_file->f_dentry->d_inode;
2004 out += sprintf(out, "%d:%s ", id, pfx);
2006 out += sprintf(out, "%6s %s ",
2007 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2008 (inode == NULL) ? "*NOINODE*" :
2009 (IS_MANDLOCK(inode) &&
2010 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2011 "MANDATORY" : "ADVISORY ");
2012 } else if (IS_FLOCK(fl)) {
2013 if (fl->fl_type & LOCK_MAND) {
2014 out += sprintf(out, "FLOCK MSNFS ");
2016 out += sprintf(out, "FLOCK ADVISORY ");
2018 } else if (IS_LEASE(fl)) {
2019 out += sprintf(out, "LEASE ");
2020 if (fl->fl_type & F_INPROGRESS)
2021 out += sprintf(out, "BREAKING ");
2022 else if (fl->fl_file)
2023 out += sprintf(out, "ACTIVE ");
2025 out += sprintf(out, "BREAKER ");
2027 out += sprintf(out, "UNKNOWN UNKNOWN ");
2029 if (fl->fl_type & LOCK_MAND) {
2030 out += sprintf(out, "%s ",
2031 (fl->fl_type & LOCK_READ)
2032 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2033 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2035 out += sprintf(out, "%s ",
2036 (fl->fl_type & F_INPROGRESS)
2037 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2038 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2041 #ifdef WE_CAN_BREAK_LSLK_NOW
2042 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2043 inode->i_sb->s_id, inode->i_ino);
2045 /* userspace relies on this representation of dev_t ;-( */
2046 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2047 MAJOR(inode->i_sb->s_dev),
2048 MINOR(inode->i_sb->s_dev), inode->i_ino);
2051 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2054 if (fl->fl_end == OFFSET_MAX)
2055 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2057 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2060 out += sprintf(out, "0 EOF\n");
2064 static void move_lock_status(char **p, off_t* pos, off_t offset)
2068 if(*pos >= offset) {
2069 /* the complete line is valid */
2074 if(*pos+len > offset) {
2075 /* use the second part of the line */
2076 int i = offset-*pos;
2077 memmove(*p,*p+i,len-i);
2082 /* discard the complete line */
2087 * get_locks_status - reports lock usage in /proc/locks
2088 * @buffer: address in userspace to write into
2090 * @offset: how far we are through the buffer
2091 * @length: how much to read
2094 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2096 struct list_head *tmp;
2102 list_for_each(tmp, &file_lock_list) {
2103 struct list_head *btmp;
2104 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2105 lock_get_status(q, fl, ++i, "");
2106 move_lock_status(&q, &pos, offset);
2108 if(pos >= offset+length)
2111 list_for_each(btmp, &fl->fl_block) {
2112 struct file_lock *bfl = list_entry(btmp,
2113 struct file_lock, fl_block);
2114 lock_get_status(q, bfl, i, " ->");
2115 move_lock_status(&q, &pos, offset);
2117 if(pos >= offset+length)
2124 if(q-buffer < length)
2130 * lock_may_read - checks that the region is free of locks
2131 * @inode: the inode that is being read
2132 * @start: the first byte to read
2133 * @len: the number of bytes to read
2135 * Emulates Windows locking requirements. Whole-file
2136 * mandatory locks (share modes) can prohibit a read and
2137 * byte-range POSIX locks can prohibit a read if they overlap.
2139 * N.B. this function is only ever called
2140 * from knfsd and ownership of locks is never checked.
2142 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2144 struct file_lock *fl;
2147 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2149 if (fl->fl_type == F_RDLCK)
2151 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2153 } else if (IS_FLOCK(fl)) {
2154 if (!(fl->fl_type & LOCK_MAND))
2156 if (fl->fl_type & LOCK_READ)
2167 EXPORT_SYMBOL(lock_may_read);
2170 * lock_may_write - checks that the region is free of locks
2171 * @inode: the inode that is being written
2172 * @start: the first byte to write
2173 * @len: the number of bytes to write
2175 * Emulates Windows locking requirements. Whole-file
2176 * mandatory locks (share modes) can prohibit a write and
2177 * byte-range POSIX locks can prohibit a write if they overlap.
2179 * N.B. this function is only ever called
2180 * from knfsd and ownership of locks is never checked.
2182 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2184 struct file_lock *fl;
2187 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2189 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2191 } else if (IS_FLOCK(fl)) {
2192 if (!(fl->fl_type & LOCK_MAND))
2194 if (fl->fl_type & LOCK_WRITE)
2205 EXPORT_SYMBOL(lock_may_write);
2207 static inline void __steal_locks(struct file *file, fl_owner_t from)
2209 struct inode *inode = file->f_dentry->d_inode;
2210 struct file_lock *fl = inode->i_flock;
2213 if (fl->fl_file == file && fl->fl_owner == from)
2214 fl->fl_owner = current->files;
2219 /* When getting ready for executing a binary, we make sure that current
2220 * has a files_struct on its own. Before dropping the old files_struct,
2221 * we take over ownership of all locks for all file descriptors we own.
2222 * Note that we may accidentally steal a lock for a file that a sibling
2223 * has created since the unshare_files() call.
2225 void steal_locks(fl_owner_t from)
2227 struct files_struct *files = current->files;
2229 struct fdtable *fdt;
2238 * We are not taking a ref to the file structures, so
2239 * we need to acquire ->file_lock.
2241 spin_lock(&files->file_lock);
2242 fdt = files_fdtable(files);
2246 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2248 set = fdt->open_fds->fds_bits[j++];
2251 struct file *file = fdt->fd[i];
2253 __steal_locks(file, from);
2259 spin_unlock(&files->file_lock);
2262 EXPORT_SYMBOL(steal_locks);
2264 static int __init filelock_init(void)
2266 filelock_cache = kmem_cache_create("file_lock_cache",
2267 sizeof(struct file_lock), 0, SLAB_PANIC,
2272 core_initcall(filelock_init);