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>
128 #include <linux/vs_limit.h>
130 #include <asm/semaphore.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable = 1;
138 int lease_break_time = 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 LIST_HEAD(file_lock_list);
145 EXPORT_SYMBOL(file_lock_list);
147 static LIST_HEAD(blocked_list);
149 static kmem_cache_t *filelock_cache;
151 /* Allocate an empty lock structure. */
152 static struct file_lock *locks_alloc_lock(void)
154 if (!vx_locks_avail(1))
156 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
159 /* Free a lock which is not in use. */
160 static void locks_free_lock(struct file_lock *fl)
169 if (waitqueue_active(&fl->fl_wait))
170 panic("Attempting to free lock with active wait queue");
172 if (!list_empty(&fl->fl_block))
173 panic("Attempting to free lock with active block list");
175 if (!list_empty(&fl->fl_link))
176 panic("Attempting to free lock on active lock list");
179 if (fl->fl_ops->fl_release_private)
180 fl->fl_ops->fl_release_private(fl);
185 if (fl->fl_lmops->fl_release_private)
186 fl->fl_lmops->fl_release_private(fl);
190 kmem_cache_free(filelock_cache, fl);
193 void locks_init_lock(struct file_lock *fl)
195 INIT_LIST_HEAD(&fl->fl_link);
196 INIT_LIST_HEAD(&fl->fl_block);
197 init_waitqueue_head(&fl->fl_wait);
199 fl->fl_fasync = NULL;
205 fl->fl_start = fl->fl_end = 0;
211 EXPORT_SYMBOL(locks_init_lock);
214 * Initialises the fields of the file lock which are invariant for
217 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
219 struct file_lock *lock = (struct file_lock *) foo;
221 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
222 SLAB_CTOR_CONSTRUCTOR)
225 locks_init_lock(lock);
229 * Initialize a new lock from an existing file_lock structure.
231 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
233 new->fl_owner = fl->fl_owner;
234 new->fl_pid = fl->fl_pid;
235 new->fl_file = fl->fl_file;
236 new->fl_flags = fl->fl_flags;
237 new->fl_type = fl->fl_type;
238 new->fl_start = fl->fl_start;
239 new->fl_end = fl->fl_end;
240 new->fl_ops = fl->fl_ops;
241 new->fl_lmops = fl->fl_lmops;
242 if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
243 fl->fl_ops->fl_copy_lock(new, fl);
244 if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
245 fl->fl_lmops->fl_copy_lock(new, fl);
247 new->fl_xid = fl->fl_xid;
250 EXPORT_SYMBOL(locks_copy_lock);
252 static inline int flock_translate_cmd(int cmd) {
254 return cmd & (LOCK_MAND | LOCK_RW);
266 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
267 static int flock_make_lock(struct file *filp, struct file_lock **lock,
270 struct file_lock *fl;
271 int type = flock_translate_cmd(cmd);
275 fl = locks_alloc_lock();
280 fl->fl_pid = current->tgid;
281 fl->fl_flags = FL_FLOCK;
283 fl->fl_end = OFFSET_MAX;
285 fl->fl_xid = filp->f_xid;
292 static int assign_type(struct file_lock *fl, int type)
306 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
309 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
314 switch (l->l_whence) {
322 start = i_size_read(filp->f_dentry->d_inode);
328 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
329 POSIX-2001 defines it. */
333 fl->fl_end = OFFSET_MAX;
335 end = start + l->l_len - 1;
337 } else if (l->l_len < 0) {
344 fl->fl_start = start; /* we record the absolute position */
345 if (fl->fl_end < fl->fl_start)
348 fl->fl_owner = current->files;
349 fl->fl_pid = current->tgid;
351 fl->fl_flags = FL_POSIX;
355 return assign_type(fl, l->l_type);
358 #if BITS_PER_LONG == 32
359 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
364 switch (l->l_whence) {
372 start = i_size_read(filp->f_dentry->d_inode);
381 fl->fl_end = OFFSET_MAX;
383 fl->fl_end = start + l->l_len - 1;
384 } else if (l->l_len < 0) {
385 fl->fl_end = start - 1;
390 fl->fl_start = start; /* we record the absolute position */
391 if (fl->fl_end < fl->fl_start)
394 fl->fl_owner = current->files;
395 fl->fl_pid = current->tgid;
397 fl->fl_flags = FL_POSIX;
405 fl->fl_type = l->l_type;
415 /* default lease lock manager operations */
416 static void lease_break_callback(struct file_lock *fl)
418 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
421 static void lease_release_private_callback(struct file_lock *fl)
426 f_delown(fl->fl_file);
427 fl->fl_file->f_owner.signum = 0;
430 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
432 return fl->fl_file == try->fl_file;
435 static struct lock_manager_operations lease_manager_ops = {
436 .fl_break = lease_break_callback,
437 .fl_release_private = lease_release_private_callback,
438 .fl_mylease = lease_mylease_callback,
439 .fl_change = lease_modify,
443 * Initialize a lease, use the default lock manager operations
445 static int lease_init(struct file *filp, int type, struct file_lock *fl)
447 if (assign_type(fl, type) != 0)
450 fl->fl_owner = current->files;
451 fl->fl_pid = current->tgid;
452 fl->fl_xid = vx_current_xid();
455 fl->fl_flags = FL_LEASE;
457 fl->fl_end = OFFSET_MAX;
459 fl->fl_lmops = &lease_manager_ops;
463 /* Allocate a file_lock initialised to this type of lease */
464 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
466 struct file_lock *fl = locks_alloc_lock();
472 fl->fl_xid = vx_current_xid();
474 error = lease_init(filp, type, fl);
484 /* Check if two locks overlap each other.
486 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
488 return ((fl1->fl_end >= fl2->fl_start) &&
489 (fl2->fl_end >= fl1->fl_start));
493 * Check whether two locks have the same owner.
495 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
497 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
498 return fl2->fl_lmops == fl1->fl_lmops &&
499 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
500 return fl1->fl_owner == fl2->fl_owner;
503 /* Remove waiter from blocker's block list.
504 * When blocker ends up pointing to itself then the list is empty.
506 static void __locks_delete_block(struct file_lock *waiter)
508 list_del_init(&waiter->fl_block);
509 list_del_init(&waiter->fl_link);
510 waiter->fl_next = NULL;
515 static void locks_delete_block(struct file_lock *waiter)
518 __locks_delete_block(waiter);
522 /* Insert waiter into blocker's block list.
523 * We use a circular list so that processes can be easily woken up in
524 * the order they blocked. The documentation doesn't require this but
525 * it seems like the reasonable thing to do.
527 static void locks_insert_block(struct file_lock *blocker,
528 struct file_lock *waiter)
530 if (!list_empty(&waiter->fl_block)) {
531 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
532 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
533 waiter->fl_start, waiter->fl_end, waiter->fl_type);
534 __locks_delete_block(waiter);
536 list_add_tail(&waiter->fl_block, &blocker->fl_block);
537 waiter->fl_next = blocker;
538 if (IS_POSIX(blocker))
539 list_add(&waiter->fl_link, &blocked_list);
542 /* Wake up processes blocked waiting for blocker.
543 * If told to wait then schedule the processes until the block list
544 * is empty, otherwise empty the block list ourselves.
546 static void locks_wake_up_blocks(struct file_lock *blocker)
548 while (!list_empty(&blocker->fl_block)) {
549 struct file_lock *waiter = list_entry(blocker->fl_block.next,
550 struct file_lock, fl_block);
551 __locks_delete_block(waiter);
552 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
553 waiter->fl_lmops->fl_notify(waiter);
555 wake_up(&waiter->fl_wait);
559 /* Insert file lock fl into an inode's lock list at the position indicated
560 * by pos. At the same time add the lock to the global file lock list.
562 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
564 list_add(&fl->fl_link, &file_lock_list);
566 /* insert into file's list */
570 if (fl->fl_ops && fl->fl_ops->fl_insert)
571 fl->fl_ops->fl_insert(fl);
575 * Delete a lock and then free it.
576 * Wake up processes that are blocked waiting for this lock,
577 * notify the FS that the lock has been cleared and
578 * finally free the lock.
580 static void locks_delete_lock(struct file_lock **thisfl_p)
582 struct file_lock *fl = *thisfl_p;
584 *thisfl_p = fl->fl_next;
586 list_del_init(&fl->fl_link);
588 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
589 if (fl->fl_fasync != NULL) {
590 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
591 fl->fl_fasync = NULL;
594 if (fl->fl_ops && fl->fl_ops->fl_remove)
595 fl->fl_ops->fl_remove(fl);
597 locks_wake_up_blocks(fl);
601 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
602 * checks for shared/exclusive status of overlapping locks.
604 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
606 if (sys_fl->fl_type == F_WRLCK)
608 if (caller_fl->fl_type == F_WRLCK)
613 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
614 * checking before calling the locks_conflict().
616 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
618 /* POSIX locks owned by the same process do not conflict with
621 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
624 /* Check whether they overlap */
625 if (!locks_overlap(caller_fl, sys_fl))
628 return (locks_conflict(caller_fl, sys_fl));
631 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
632 * checking before calling the locks_conflict().
634 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
636 /* FLOCK locks referring to the same filp do not conflict with
639 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
641 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
644 return (locks_conflict(caller_fl, sys_fl));
647 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
650 DECLARE_WAITQUEUE(wait, current);
652 __set_current_state(TASK_INTERRUPTIBLE);
653 add_wait_queue(fl_wait, &wait);
657 result = schedule_timeout(timeout);
658 if (signal_pending(current))
659 result = -ERESTARTSYS;
660 remove_wait_queue(fl_wait, &wait);
661 __set_current_state(TASK_RUNNING);
665 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
668 locks_insert_block(blocker, waiter);
669 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
670 __locks_delete_block(waiter);
675 posix_test_lock(struct file *filp, struct file_lock *fl)
677 struct file_lock *cfl;
680 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
683 if (posix_locks_conflict(cfl, fl))
691 EXPORT_SYMBOL(posix_test_lock);
693 /* This function tests for deadlock condition before putting a process to
694 * sleep. The detection scheme is no longer recursive. Recursive was neat,
695 * but dangerous - we risked stack corruption if the lock data was bad, or
696 * if the recursion was too deep for any other reason.
698 * We rely on the fact that a task can only be on one lock's wait queue
699 * at a time. When we find blocked_task on a wait queue we can re-search
700 * with blocked_task equal to that queue's owner, until either blocked_task
701 * isn't found, or blocked_task is found on a queue owned by my_task.
703 * Note: the above assumption may not be true when handling lock requests
704 * from a broken NFS client. But broken NFS clients have a lot more to
705 * worry about than proper deadlock detection anyway... --okir
707 int posix_locks_deadlock(struct file_lock *caller_fl,
708 struct file_lock *block_fl)
710 struct list_head *tmp;
713 if (posix_same_owner(caller_fl, block_fl))
715 list_for_each(tmp, &blocked_list) {
716 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
717 if (posix_same_owner(fl, block_fl)) {
726 EXPORT_SYMBOL(posix_locks_deadlock);
728 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
729 * at the head of the list, but that's secret knowledge known only to
730 * flock_lock_file and posix_lock_file.
732 static int flock_lock_file(struct file *filp, struct file_lock *request)
734 struct file_lock *new_fl = NULL;
735 struct file_lock **before;
736 struct inode * inode = filp->f_dentry->d_inode;
741 for_each_lock(inode, before) {
742 struct file_lock *fl = *before;
747 if (filp != fl->fl_file)
749 if (request->fl_type == fl->fl_type)
752 locks_delete_lock(before);
756 if (request->fl_type == F_UNLCK)
759 new_fl = locks_alloc_lock();
763 * If a higher-priority process was blocked on the old file lock,
764 * give it the opportunity to lock the file.
769 for_each_lock(inode, before) {
770 struct file_lock *fl = *before;
775 if (!flock_locks_conflict(request, fl))
778 if (request->fl_flags & FL_SLEEP)
779 locks_insert_block(fl, request);
782 locks_copy_lock(new_fl, request);
783 vx_locks_inc(new_fl);
784 locks_insert_lock(&inode->i_flock, new_fl);
790 locks_free_lock(new_fl);
794 EXPORT_SYMBOL(posix_lock_file);
796 static int __posix_lock_file(struct inode *inode, struct file_lock *request, xid_t xid)
798 struct file_lock *fl;
799 struct file_lock *new_fl, *new_fl2;
800 struct file_lock *left = NULL;
801 struct file_lock *right = NULL;
802 struct file_lock **before;
803 int error, added = 0;
806 * We may need two file_lock structures for this operation,
807 * so we get them in advance to avoid races.
809 new_fl = locks_alloc_lock();
810 new_fl->fl_xid = xid;
811 vx_locks_inc(new_fl);
812 new_fl2 = locks_alloc_lock();
813 new_fl2->fl_xid = xid;
814 vx_locks_inc(new_fl2);
817 if (request->fl_type != F_UNLCK) {
818 for_each_lock(inode, before) {
819 struct file_lock *fl = *before;
822 if (!posix_locks_conflict(request, fl))
825 if (!(request->fl_flags & FL_SLEEP))
828 if (posix_locks_deadlock(request, fl))
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)
841 error = -ENOLCK; /* "no luck" */
842 if (!(new_fl && new_fl2))
846 * We've allocated the new locks in advance, so there are no
847 * errors possible (and no blocking operations) from here on.
849 * Find the first old lock with the same owner as the new lock.
852 before = &inode->i_flock;
854 /* First skip locks owned by other processes. */
855 while ((fl = *before) && (!IS_POSIX(fl) ||
856 !posix_same_owner(request, fl))) {
857 before = &fl->fl_next;
860 /* Process locks with this owner. */
861 while ((fl = *before) && posix_same_owner(request, fl)) {
862 /* Detect adjacent or overlapping regions (if same lock type)
864 if (request->fl_type == fl->fl_type) {
865 /* In all comparisons of start vs end, use
866 * "start - 1" rather than "end + 1". If end
867 * is OFFSET_MAX, end + 1 will become negative.
869 if (fl->fl_end < request->fl_start - 1)
871 /* If the next lock in the list has entirely bigger
872 * addresses than the new one, insert the lock here.
874 if (fl->fl_start - 1 > request->fl_end)
877 /* If we come here, the new and old lock are of the
878 * same type and adjacent or overlapping. Make one
879 * lock yielding from the lower start address of both
880 * locks to the higher end address.
882 if (fl->fl_start > request->fl_start)
883 fl->fl_start = request->fl_start;
885 request->fl_start = fl->fl_start;
886 if (fl->fl_end < request->fl_end)
887 fl->fl_end = request->fl_end;
889 request->fl_end = fl->fl_end;
891 locks_delete_lock(before);
898 /* Processing for different lock types is a bit
901 if (fl->fl_end < request->fl_start)
903 if (fl->fl_start > request->fl_end)
905 if (request->fl_type == F_UNLCK)
907 if (fl->fl_start < request->fl_start)
909 /* If the next lock in the list has a higher end
910 * address than the new one, insert the new one here.
912 if (fl->fl_end > request->fl_end) {
916 if (fl->fl_start >= request->fl_start) {
917 /* The new lock completely replaces an old
918 * one (This may happen several times).
921 locks_delete_lock(before);
924 /* Replace the old lock with the new one.
925 * Wake up anybody waiting for the old one,
926 * as the change in lock type might satisfy
929 locks_wake_up_blocks(fl);
930 fl->fl_start = request->fl_start;
931 fl->fl_end = request->fl_end;
932 fl->fl_type = request->fl_type;
933 fl->fl_u = request->fl_u;
938 /* Go on to next lock.
941 before = &fl->fl_next;
946 if (request->fl_type == F_UNLCK)
948 locks_copy_lock(new_fl, request);
949 locks_insert_lock(before, new_fl);
954 /* The new lock breaks the old one in two pieces,
955 * so we have to use the second new lock.
959 locks_copy_lock(left, right);
960 locks_insert_lock(before, left);
962 right->fl_start = request->fl_end + 1;
963 locks_wake_up_blocks(right);
966 left->fl_end = request->fl_start - 1;
967 locks_wake_up_blocks(left);
972 * Free any unused locks.
975 locks_free_lock(new_fl);
977 locks_free_lock(new_fl2);
982 * posix_lock_file - Apply a POSIX-style lock to a file
983 * @filp: The file to apply the lock to
984 * @fl: The lock to be applied
986 * Add a POSIX style lock to a file.
987 * We merge adjacent & overlapping locks whenever possible.
988 * POSIX locks are sorted by owner task, then by starting address
990 int posix_lock_file(struct file *filp, struct file_lock *fl)
992 return __posix_lock_file(filp->f_dentry->d_inode, fl, filp->f_xid);
996 * posix_lock_file_wait - Apply a POSIX-style lock to a file
997 * @filp: The file to apply the lock to
998 * @fl: The lock to be applied
1000 * Add a POSIX style lock to a file.
1001 * We merge adjacent & overlapping locks whenever possible.
1002 * POSIX locks are sorted by owner task, then by starting address
1004 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1009 error = __posix_lock_file(filp->f_dentry->d_inode,
1011 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1013 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1017 locks_delete_block(fl);
1022 EXPORT_SYMBOL(posix_lock_file_wait);
1025 * locks_mandatory_locked - Check for an active lock
1026 * @inode: the file to check
1028 * Searches the inode's list of locks to find any POSIX locks which conflict.
1029 * This function is called from locks_verify_locked() only.
1031 int locks_mandatory_locked(struct inode *inode)
1033 fl_owner_t owner = current->files;
1034 struct file_lock *fl;
1037 * Search the lock list for this inode for any POSIX locks.
1040 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1043 if (fl->fl_owner != owner)
1047 return fl ? -EAGAIN : 0;
1051 * locks_mandatory_area - Check for a conflicting lock
1052 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1054 * @inode: the file to check
1055 * @filp: how the file was opened (if it was)
1056 * @offset: start of area to check
1057 * @count: length of area to check
1059 * Searches the inode's list of locks to find any POSIX locks which conflict.
1060 * This function is called from rw_verify_area() and
1061 * locks_verify_truncate().
1063 int locks_mandatory_area(int read_write, struct inode *inode,
1064 struct file *filp, loff_t offset,
1067 struct file_lock fl;
1070 locks_init_lock(&fl);
1071 fl.fl_owner = current->files;
1072 fl.fl_pid = current->tgid;
1074 fl.fl_flags = FL_POSIX | FL_ACCESS;
1075 if (filp && !(filp->f_flags & O_NONBLOCK))
1076 fl.fl_flags |= FL_SLEEP;
1077 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1078 fl.fl_start = offset;
1079 fl.fl_end = offset + count - 1;
1082 error = __posix_lock_file(inode, &fl, filp->f_xid);
1083 if (error != -EAGAIN)
1085 if (!(fl.fl_flags & FL_SLEEP))
1087 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1090 * If we've been sleeping someone might have
1091 * changed the permissions behind our back.
1093 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1097 locks_delete_block(&fl);
1104 EXPORT_SYMBOL(locks_mandatory_area);
1106 /* We already had a lease on this file; just change its type */
1107 int lease_modify(struct file_lock **before, int arg)
1109 struct file_lock *fl = *before;
1110 int error = assign_type(fl, arg);
1114 locks_wake_up_blocks(fl);
1116 locks_delete_lock(before);
1120 EXPORT_SYMBOL(lease_modify);
1122 static void time_out_leases(struct inode *inode)
1124 struct file_lock **before;
1125 struct file_lock *fl;
1127 before = &inode->i_flock;
1128 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1129 if ((fl->fl_break_time == 0)
1130 || time_before(jiffies, fl->fl_break_time)) {
1131 before = &fl->fl_next;
1134 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1135 if (fl == *before) /* lease_modify may have freed fl */
1136 before = &fl->fl_next;
1141 * __break_lease - revoke all outstanding leases on file
1142 * @inode: the inode of the file to return
1143 * @mode: the open mode (read or write)
1145 * break_lease (inlined for speed) has checked there already
1146 * is a lease on this file. Leases are broken on a call to open()
1147 * or truncate(). This function can sleep unless you
1148 * specified %O_NONBLOCK to your open().
1150 int __break_lease(struct inode *inode, unsigned int mode)
1152 int error = 0, future;
1153 struct file_lock *new_fl, *flock;
1154 struct file_lock *fl;
1156 unsigned long break_time;
1157 int i_have_this_lease = 0;
1159 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1164 time_out_leases(inode);
1166 flock = inode->i_flock;
1167 if ((flock == NULL) || !IS_LEASE(flock))
1170 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1171 if (fl->fl_owner == current->files)
1172 i_have_this_lease = 1;
1174 if (mode & FMODE_WRITE) {
1175 /* If we want write access, we have to revoke any lease. */
1176 future = F_UNLCK | F_INPROGRESS;
1177 } else if (flock->fl_type & F_INPROGRESS) {
1178 /* If the lease is already being broken, we just leave it */
1179 future = flock->fl_type;
1180 } else if (flock->fl_type & F_WRLCK) {
1181 /* Downgrade the exclusive lease to a read-only lease. */
1182 future = F_RDLCK | F_INPROGRESS;
1184 /* the existing lease was read-only, so we can read too. */
1188 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1194 if (lease_break_time > 0) {
1195 break_time = jiffies + lease_break_time * HZ;
1196 if (break_time == 0)
1197 break_time++; /* so that 0 means no break time */
1200 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1201 if (fl->fl_type != future) {
1202 fl->fl_type = future;
1203 fl->fl_break_time = break_time;
1204 /* lease must have lmops break callback */
1205 fl->fl_lmops->fl_break(fl);
1209 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1210 error = -EWOULDBLOCK;
1215 break_time = flock->fl_break_time;
1216 if (break_time != 0) {
1217 break_time -= jiffies;
1218 if (break_time == 0)
1221 error = locks_block_on_timeout(flock, new_fl, break_time);
1224 time_out_leases(inode);
1225 /* Wait for the next lease that has not been broken yet */
1226 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1227 flock = flock->fl_next) {
1228 if (flock->fl_type & F_INPROGRESS)
1237 locks_free_lock(new_fl);
1241 EXPORT_SYMBOL(__break_lease);
1246 * @time: pointer to a timespec which will contain the last modified time
1248 * This is to force NFS clients to flush their caches for files with
1249 * exclusive leases. The justification is that if someone has an
1250 * exclusive lease, then they could be modifiying it.
1252 void lease_get_mtime(struct inode *inode, struct timespec *time)
1254 struct file_lock *flock = inode->i_flock;
1255 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1256 *time = current_fs_time(inode->i_sb);
1258 *time = inode->i_mtime;
1261 EXPORT_SYMBOL(lease_get_mtime);
1264 * fcntl_getlease - Enquire what lease is currently active
1267 * The value returned by this function will be one of
1268 * (if no lease break is pending):
1270 * %F_RDLCK to indicate a shared lease is held.
1272 * %F_WRLCK to indicate an exclusive lease is held.
1274 * %F_UNLCK to indicate no lease is held.
1276 * (if a lease break is pending):
1278 * %F_RDLCK to indicate an exclusive lease needs to be
1279 * changed to a shared lease (or removed).
1281 * %F_UNLCK to indicate the lease needs to be removed.
1283 * XXX: sfr & willy disagree over whether F_INPROGRESS
1284 * should be returned to userspace.
1286 int fcntl_getlease(struct file *filp)
1288 struct file_lock *fl;
1292 time_out_leases(filp->f_dentry->d_inode);
1293 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1295 if (fl->fl_file == filp) {
1296 type = fl->fl_type & ~F_INPROGRESS;
1305 * __setlease - sets a lease on an open file
1306 * @filp: file pointer
1307 * @arg: type of lease to obtain
1308 * @flp: input - file_lock to use, output - file_lock inserted
1310 * The (input) flp->fl_lmops->fl_break function is required
1313 * Called with kernel lock held.
1315 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1317 struct file_lock *fl, **before, **my_before = NULL, *lease;
1318 struct dentry *dentry = filp->f_dentry;
1319 struct inode *inode = dentry->d_inode;
1320 int error, rdlease_count = 0, wrlease_count = 0;
1322 time_out_leases(inode);
1325 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1331 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1333 if ((arg == F_WRLCK)
1334 && ((atomic_read(&dentry->d_count) > 1)
1335 || (atomic_read(&inode->i_count) > 1)))
1339 * At this point, we know that if there is an exclusive
1340 * lease on this file, then we hold it on this filp
1341 * (otherwise our open of this file would have blocked).
1342 * And if we are trying to acquire an exclusive lease,
1343 * then the file is not open by anyone (including us)
1344 * except for this filp.
1346 for (before = &inode->i_flock;
1347 ((fl = *before) != NULL) && IS_LEASE(fl);
1348 before = &fl->fl_next) {
1349 if (lease->fl_lmops->fl_mylease(fl, lease))
1351 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1353 * Someone is in the process of opening this
1354 * file for writing so we may not take an
1355 * exclusive lease on it.
1362 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1363 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1366 if (my_before != NULL) {
1368 error = lease->fl_lmops->fl_change(my_before, arg);
1381 fl = locks_alloc_lock();
1385 locks_copy_lock(fl, lease);
1387 locks_insert_lock(before, fl);
1396 * setlease - sets a lease on an open file
1397 * @filp: file pointer
1398 * @arg: type of lease to obtain
1399 * @lease: file_lock to use
1401 * Call this to establish a lease on the file.
1402 * The fl_lmops fl_break function is required by break_lease
1405 int setlease(struct file *filp, long arg, struct file_lock **lease)
1407 struct dentry *dentry = filp->f_dentry;
1408 struct inode *inode = dentry->d_inode;
1411 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1413 if (!S_ISREG(inode->i_mode))
1415 error = security_file_lock(filp, arg);
1420 error = __setlease(filp, arg, lease);
1426 EXPORT_SYMBOL(setlease);
1429 * fcntl_setlease - sets a lease on an open file
1430 * @fd: open file descriptor
1431 * @filp: file pointer
1432 * @arg: type of lease to obtain
1434 * Call this fcntl to establish a lease on the file.
1435 * Note that you also need to call %F_SETSIG to
1436 * receive a signal when the lease is broken.
1438 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1440 struct file_lock fl, *flp = &fl;
1441 struct dentry *dentry = filp->f_dentry;
1442 struct inode *inode = dentry->d_inode;
1445 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1447 if (!S_ISREG(inode->i_mode))
1449 error = security_file_lock(filp, arg);
1453 locks_init_lock(&fl);
1454 error = lease_init(filp, arg, &fl);
1460 error = __setlease(filp, arg, &flp);
1461 if (error || arg == F_UNLCK)
1464 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1466 /* remove lease just inserted by __setlease */
1467 flp->fl_type = F_UNLCK | F_INPROGRESS;
1468 flp->fl_break_time = jiffies- 10;
1469 time_out_leases(inode);
1473 error = f_setown(filp, current->pid, 0);
1480 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1481 * @filp: The file to apply the lock to
1482 * @fl: The lock to be applied
1484 * Add a FLOCK style lock to a file.
1486 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1491 error = flock_lock_file(filp, fl);
1492 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1494 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1498 locks_delete_block(fl);
1504 EXPORT_SYMBOL(flock_lock_file_wait);
1507 * sys_flock: - flock() system call.
1508 * @fd: the file descriptor to lock.
1509 * @cmd: the type of lock to apply.
1511 * Apply a %FL_FLOCK style lock to an open file descriptor.
1512 * The @cmd can be one of
1514 * %LOCK_SH -- a shared lock.
1516 * %LOCK_EX -- an exclusive lock.
1518 * %LOCK_UN -- remove an existing lock.
1520 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1522 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1523 * processes read and write access respectively.
1525 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1528 struct file_lock *lock;
1529 int can_sleep, unlock;
1537 can_sleep = !(cmd & LOCK_NB);
1539 unlock = (cmd == LOCK_UN);
1541 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1544 error = flock_make_lock(filp, &lock, cmd);
1548 lock->fl_flags |= FL_SLEEP;
1550 error = security_file_lock(filp, cmd);
1554 if (filp->f_op && filp->f_op->flock)
1555 error = filp->f_op->flock(filp,
1556 (can_sleep) ? F_SETLKW : F_SETLK,
1559 error = flock_lock_file_wait(filp, lock);
1562 locks_free_lock(lock);
1570 /* Report the first existing lock that would conflict with l.
1571 * This implements the F_GETLK command of fcntl().
1573 int fcntl_getlk(struct file *filp, struct flock __user *l)
1575 struct file_lock *fl, file_lock;
1580 if (copy_from_user(&flock, l, sizeof(flock)))
1583 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1586 error = flock_to_posix_lock(filp, &file_lock, &flock);
1590 if (filp->f_op && filp->f_op->lock) {
1591 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1592 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1593 file_lock.fl_ops->fl_release_private(&file_lock);
1597 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1599 fl = posix_test_lock(filp, &file_lock);
1602 flock.l_type = F_UNLCK;
1604 flock.l_pid = fl->fl_pid;
1605 #if BITS_PER_LONG == 32
1607 * Make sure we can represent the posix lock via
1608 * legacy 32bit flock.
1611 if (fl->fl_start > OFFT_OFFSET_MAX)
1613 if ((fl->fl_end != OFFSET_MAX)
1614 && (fl->fl_end > OFFT_OFFSET_MAX))
1617 flock.l_start = fl->fl_start;
1618 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1619 fl->fl_end - fl->fl_start + 1;
1621 flock.l_type = fl->fl_type;
1624 if (!copy_to_user(l, &flock, sizeof(flock)))
1630 /* Apply the lock described by l to an open file descriptor.
1631 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1633 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1634 struct flock __user *l)
1636 struct file_lock *file_lock = locks_alloc_lock();
1638 struct inode *inode;
1641 if (file_lock == NULL)
1644 file_lock->fl_xid = filp->f_xid;
1645 vx_locks_inc(file_lock);
1648 * This might block, so we do it before checking the inode.
1651 if (copy_from_user(&flock, l, sizeof(flock)))
1654 inode = filp->f_dentry->d_inode;
1656 /* Don't allow mandatory locks on files that may be memory mapped
1659 if (IS_MANDLOCK(inode) &&
1660 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1661 mapping_writably_mapped(filp->f_mapping)) {
1667 error = flock_to_posix_lock(filp, file_lock, &flock);
1670 if (cmd == F_SETLKW) {
1671 file_lock->fl_flags |= FL_SLEEP;
1675 switch (flock.l_type) {
1677 if (!(filp->f_mode & FMODE_READ))
1681 if (!(filp->f_mode & FMODE_WRITE))
1691 error = security_file_lock(filp, file_lock->fl_type);
1695 if (filp->f_op && filp->f_op->lock != NULL)
1696 error = filp->f_op->lock(filp, cmd, file_lock);
1699 error = __posix_lock_file(inode, file_lock,
1701 if ((error != -EAGAIN) || (cmd == F_SETLK))
1703 error = wait_event_interruptible(file_lock->fl_wait,
1704 !file_lock->fl_next);
1708 locks_delete_block(file_lock);
1714 * Attempt to detect a close/fcntl race and recover by
1715 * releasing the lock that was just acquired.
1717 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1718 flock.l_type = F_UNLCK;
1723 locks_free_lock(file_lock);
1727 #if BITS_PER_LONG == 32
1728 /* Report the first existing lock that would conflict with l.
1729 * This implements the F_GETLK command of fcntl().
1731 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1733 struct file_lock *fl, file_lock;
1734 struct flock64 flock;
1738 if (copy_from_user(&flock, l, sizeof(flock)))
1741 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1744 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1748 if (filp->f_op && filp->f_op->lock) {
1749 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1750 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1751 file_lock.fl_ops->fl_release_private(&file_lock);
1755 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1757 fl = posix_test_lock(filp, &file_lock);
1760 flock.l_type = F_UNLCK;
1762 flock.l_pid = fl->fl_pid;
1763 flock.l_start = fl->fl_start;
1764 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1765 fl->fl_end - fl->fl_start + 1;
1767 flock.l_type = fl->fl_type;
1770 if (!copy_to_user(l, &flock, sizeof(flock)))
1777 /* Apply the lock described by l to an open file descriptor.
1778 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1780 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1781 struct flock64 __user *l)
1783 struct file_lock *file_lock = locks_alloc_lock();
1784 struct flock64 flock;
1785 struct inode *inode;
1788 if (file_lock == NULL)
1791 file_lock->fl_xid = filp->f_xid;
1792 vx_locks_inc(file_lock);
1795 * This might block, so we do it before checking the inode.
1798 if (copy_from_user(&flock, l, sizeof(flock)))
1801 inode = filp->f_dentry->d_inode;
1803 /* Don't allow mandatory locks on files that may be memory mapped
1806 if (IS_MANDLOCK(inode) &&
1807 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1808 mapping_writably_mapped(filp->f_mapping)) {
1814 error = flock64_to_posix_lock(filp, file_lock, &flock);
1817 if (cmd == F_SETLKW64) {
1818 file_lock->fl_flags |= FL_SLEEP;
1822 switch (flock.l_type) {
1824 if (!(filp->f_mode & FMODE_READ))
1828 if (!(filp->f_mode & FMODE_WRITE))
1838 error = security_file_lock(filp, file_lock->fl_type);
1842 if (filp->f_op && filp->f_op->lock != NULL)
1843 error = filp->f_op->lock(filp, cmd, file_lock);
1846 error = __posix_lock_file(inode, 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_block_lock - blocks waiting for a file lock
1976 * @blocker: the lock which is blocking
1977 * @waiter: the lock which conflicts and has to wait
1979 * lockd needs to block waiting for locks.
1982 posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1984 locks_insert_block(blocker, waiter);
1987 EXPORT_SYMBOL(posix_block_lock);
1990 * posix_unblock_lock - stop waiting for a file lock
1991 * @filp: how the file was opened
1992 * @waiter: the lock which was waiting
1994 * lockd needs to block waiting for locks.
1997 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2002 if (waiter->fl_next)
2003 __locks_delete_block(waiter);
2010 EXPORT_SYMBOL(posix_unblock_lock);
2012 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2014 struct inode *inode = NULL;
2016 if (fl->fl_file != NULL)
2017 inode = fl->fl_file->f_dentry->d_inode;
2019 out += sprintf(out, "%d:%s ", id, pfx);
2021 out += sprintf(out, "%6s %s ",
2022 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2023 (inode == NULL) ? "*NOINODE*" :
2024 (IS_MANDLOCK(inode) &&
2025 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2026 "MANDATORY" : "ADVISORY ");
2027 } else if (IS_FLOCK(fl)) {
2028 if (fl->fl_type & LOCK_MAND) {
2029 out += sprintf(out, "FLOCK MSNFS ");
2031 out += sprintf(out, "FLOCK ADVISORY ");
2033 } else if (IS_LEASE(fl)) {
2034 out += sprintf(out, "LEASE ");
2035 if (fl->fl_type & F_INPROGRESS)
2036 out += sprintf(out, "BREAKING ");
2037 else if (fl->fl_file)
2038 out += sprintf(out, "ACTIVE ");
2040 out += sprintf(out, "BREAKER ");
2042 out += sprintf(out, "UNKNOWN UNKNOWN ");
2044 if (fl->fl_type & LOCK_MAND) {
2045 out += sprintf(out, "%s ",
2046 (fl->fl_type & LOCK_READ)
2047 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2048 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2050 out += sprintf(out, "%s ",
2051 (fl->fl_type & F_INPROGRESS)
2052 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2053 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2056 #ifdef WE_CAN_BREAK_LSLK_NOW
2057 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2058 inode->i_sb->s_id, inode->i_ino);
2060 /* userspace relies on this representation of dev_t ;-( */
2061 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2062 MAJOR(inode->i_sb->s_dev),
2063 MINOR(inode->i_sb->s_dev), inode->i_ino);
2066 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2069 if (fl->fl_end == OFFSET_MAX)
2070 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2072 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2075 out += sprintf(out, "0 EOF\n");
2079 static void move_lock_status(char **p, off_t* pos, off_t offset)
2083 if(*pos >= offset) {
2084 /* the complete line is valid */
2089 if(*pos+len > offset) {
2090 /* use the second part of the line */
2091 int i = offset-*pos;
2092 memmove(*p,*p+i,len-i);
2097 /* discard the complete line */
2102 * get_locks_status - reports lock usage in /proc/locks
2103 * @buffer: address in userspace to write into
2105 * @offset: how far we are through the buffer
2106 * @length: how much to read
2109 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2111 struct list_head *tmp;
2117 list_for_each(tmp, &file_lock_list) {
2118 struct list_head *btmp;
2119 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2121 if (!vx_check(fl->fl_xid, VX_IDENT|VX_WATCH))
2124 lock_get_status(q, fl, ++i, "");
2125 move_lock_status(&q, &pos, offset);
2127 if(pos >= offset+length)
2130 list_for_each(btmp, &fl->fl_block) {
2131 struct file_lock *bfl = list_entry(btmp,
2132 struct file_lock, fl_block);
2133 lock_get_status(q, bfl, i, " ->");
2134 move_lock_status(&q, &pos, offset);
2136 if(pos >= offset+length)
2143 if(q-buffer < length)
2149 * lock_may_read - checks that the region is free of locks
2150 * @inode: the inode that is being read
2151 * @start: the first byte to read
2152 * @len: the number of bytes to read
2154 * Emulates Windows locking requirements. Whole-file
2155 * mandatory locks (share modes) can prohibit a read and
2156 * byte-range POSIX locks can prohibit a read if they overlap.
2158 * N.B. this function is only ever called
2159 * from knfsd and ownership of locks is never checked.
2161 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2163 struct file_lock *fl;
2166 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2168 if (fl->fl_type == F_RDLCK)
2170 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2172 } else if (IS_FLOCK(fl)) {
2173 if (!(fl->fl_type & LOCK_MAND))
2175 if (fl->fl_type & LOCK_READ)
2186 EXPORT_SYMBOL(lock_may_read);
2189 * lock_may_write - checks that the region is free of locks
2190 * @inode: the inode that is being written
2191 * @start: the first byte to write
2192 * @len: the number of bytes to write
2194 * Emulates Windows locking requirements. Whole-file
2195 * mandatory locks (share modes) can prohibit a write and
2196 * byte-range POSIX locks can prohibit a write if they overlap.
2198 * N.B. this function is only ever called
2199 * from knfsd and ownership of locks is never checked.
2201 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2203 struct file_lock *fl;
2206 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2208 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2210 } else if (IS_FLOCK(fl)) {
2211 if (!(fl->fl_type & LOCK_MAND))
2213 if (fl->fl_type & LOCK_WRITE)
2224 EXPORT_SYMBOL(lock_may_write);
2226 static inline void __steal_locks(struct file *file, fl_owner_t from)
2228 struct inode *inode = file->f_dentry->d_inode;
2229 struct file_lock *fl = inode->i_flock;
2232 if (fl->fl_file == file && fl->fl_owner == from)
2233 fl->fl_owner = current->files;
2238 /* When getting ready for executing a binary, we make sure that current
2239 * has a files_struct on its own. Before dropping the old files_struct,
2240 * we take over ownership of all locks for all file descriptors we own.
2241 * Note that we may accidentally steal a lock for a file that a sibling
2242 * has created since the unshare_files() call.
2244 void steal_locks(fl_owner_t from)
2246 struct files_struct *files = current->files;
2248 struct fdtable *fdt;
2257 * We are not taking a ref to the file structures, so
2258 * we need to acquire ->file_lock.
2260 spin_lock(&files->file_lock);
2261 fdt = files_fdtable(files);
2265 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2267 set = fdt->open_fds->fds_bits[j++];
2270 struct file *file = fdt->fd[i];
2272 __steal_locks(file, from);
2278 spin_unlock(&files->file_lock);
2281 EXPORT_SYMBOL(steal_locks);
2283 static int __init filelock_init(void)
2285 filelock_cache = kmem_cache_create("file_lock_cache",
2286 sizeof(struct file_lock), 0, SLAB_PANIC,
2291 core_initcall(filelock_init);