]> nv-tegra.nvidia Code Review - linux-3.10.git/blobdiff - fs/locks.c
ext3: Fix format string issues
[linux-3.10.git] / fs / locks.c
index 4a8072736efa2716b46121bb8d6b994c7f573054..cb424a4fed71a42d237b72d16de8318bbbeb0fbd 100644 (file)
@@ -60,7 +60,7 @@
  *
  *  Initial implementation of mandatory locks. SunOS turned out to be
  *  a rotten model, so I implemented the "obvious" semantics.
- *  See 'Documentation/mandatory.txt' for details.
+ *  See 'Documentation/filesystems/mandatory-locking.txt' for details.
  *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
  *
  *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
 
 #include <linux/capability.h>
 #include <linux/file.h>
+#include <linux/fdtable.h>
 #include <linux/fs.h>
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/security.h>
 #include <linux/slab.h>
-#include <linux/smp_lock.h>
 #include <linux/syscalls.h>
 #include <linux/time.h>
 #include <linux/rcupdate.h>
+#include <linux/pid_namespace.h>
 
-#include <asm/semaphore.h>
 #include <asm/uaccess.h>
 
 #define IS_POSIX(fl)   (fl->fl_flags & FL_POSIX)
 #define IS_FLOCK(fl)   (fl->fl_flags & FL_FLOCK)
 #define IS_LEASE(fl)   (fl->fl_flags & FL_LEASE)
 
+static bool lease_breaking(struct file_lock *fl)
+{
+       return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
+}
+
+static int target_leasetype(struct file_lock *fl)
+{
+       if (fl->fl_flags & FL_UNLOCK_PENDING)
+               return F_UNLCK;
+       if (fl->fl_flags & FL_DOWNGRADE_PENDING)
+               return F_RDLCK;
+       return fl->fl_type;
+}
+
 int leases_enable = 1;
 int lease_break_time = 45;
 
@@ -141,32 +155,58 @@ int lease_break_time = 45;
 
 static LIST_HEAD(file_lock_list);
 static LIST_HEAD(blocked_list);
+static DEFINE_SPINLOCK(file_lock_lock);
+
+/*
+ * Protects the two list heads above, plus the inode->i_flock list
+ */
+void lock_flocks(void)
+{
+       spin_lock(&file_lock_lock);
+}
+EXPORT_SYMBOL_GPL(lock_flocks);
+
+void unlock_flocks(void)
+{
+       spin_unlock(&file_lock_lock);
+}
+EXPORT_SYMBOL_GPL(unlock_flocks);
 
 static struct kmem_cache *filelock_cache __read_mostly;
 
+static void locks_init_lock_heads(struct file_lock *fl)
+{
+       INIT_LIST_HEAD(&fl->fl_link);
+       INIT_LIST_HEAD(&fl->fl_block);
+       init_waitqueue_head(&fl->fl_wait);
+}
+
 /* Allocate an empty lock structure. */
-static struct file_lock *locks_alloc_lock(void)
+struct file_lock *locks_alloc_lock(void)
 {
-       return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
+       struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
+
+       if (fl)
+               locks_init_lock_heads(fl);
+
+       return fl;
 }
+EXPORT_SYMBOL_GPL(locks_alloc_lock);
 
-static void locks_release_private(struct file_lock *fl)
+void locks_release_private(struct file_lock *fl)
 {
        if (fl->fl_ops) {
                if (fl->fl_ops->fl_release_private)
                        fl->fl_ops->fl_release_private(fl);
                fl->fl_ops = NULL;
        }
-       if (fl->fl_lmops) {
-               if (fl->fl_lmops->fl_release_private)
-                       fl->fl_lmops->fl_release_private(fl);
-               fl->fl_lmops = NULL;
-       }
+       fl->fl_lmops = NULL;
 
 }
+EXPORT_SYMBOL_GPL(locks_release_private);
 
 /* Free a lock which is not in use. */
-static void locks_free_lock(struct file_lock *fl)
+void locks_free_lock(struct file_lock *fl)
 {
        BUG_ON(waitqueue_active(&fl->fl_wait));
        BUG_ON(!list_empty(&fl->fl_block));
@@ -175,37 +215,16 @@ static void locks_free_lock(struct file_lock *fl)
        locks_release_private(fl);
        kmem_cache_free(filelock_cache, fl);
 }
+EXPORT_SYMBOL(locks_free_lock);
 
 void locks_init_lock(struct file_lock *fl)
 {
-       INIT_LIST_HEAD(&fl->fl_link);
-       INIT_LIST_HEAD(&fl->fl_block);
-       init_waitqueue_head(&fl->fl_wait);
-       fl->fl_next = NULL;
-       fl->fl_fasync = NULL;
-       fl->fl_owner = NULL;
-       fl->fl_pid = 0;
-       fl->fl_file = NULL;
-       fl->fl_flags = 0;
-       fl->fl_type = 0;
-       fl->fl_start = fl->fl_end = 0;
-       fl->fl_ops = NULL;
-       fl->fl_lmops = NULL;
+       memset(fl, 0, sizeof(struct file_lock));
+       locks_init_lock_heads(fl);
 }
 
 EXPORT_SYMBOL(locks_init_lock);
 
-/*
- * Initialises the fields of the file lock which are invariant for
- * free file_locks.
- */
-static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
-{
-       struct file_lock *lock = (struct file_lock *) foo;
-
-       locks_init_lock(lock);
-}
-
 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
 {
        if (fl->fl_ops) {
@@ -213,17 +232,14 @@ static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
                        fl->fl_ops->fl_copy_lock(new, fl);
                new->fl_ops = fl->fl_ops;
        }
-       if (fl->fl_lmops) {
-               if (fl->fl_lmops->fl_copy_lock)
-                       fl->fl_lmops->fl_copy_lock(new, fl);
+       if (fl->fl_lmops)
                new->fl_lmops = fl->fl_lmops;
-       }
 }
 
 /*
  * Initialize a new lock from an existing file_lock structure.
  */
-static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
+void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
 {
        new->fl_owner = fl->fl_owner;
        new->fl_pid = fl->fl_pid;
@@ -235,6 +251,7 @@ static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
        new->fl_ops = NULL;
        new->fl_lmops = NULL;
 }
+EXPORT_SYMBOL(__locks_copy_lock);
 
 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
 {
@@ -287,7 +304,7 @@ static int flock_make_lock(struct file *filp, struct file_lock **lock,
        return 0;
 }
 
-static int assign_type(struct file_lock *fl, int type)
+static int assign_type(struct file_lock *fl, long type)
 {
        switch (type) {
        case F_RDLCK:
@@ -317,7 +334,7 @@ static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
                start = filp->f_pos;
                break;
        case SEEK_END:
-               start = i_size_read(filp->f_path.dentry->d_inode);
+               start = i_size_read(file_inode(filp));
                break;
        default:
                return -EINVAL;
@@ -367,7 +384,7 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
                start = filp->f_pos;
                break;
        case SEEK_END:
-               start = i_size_read(filp->f_path.dentry->d_inode);
+               start = i_size_read(file_inode(filp));
                break;
        default:
                return -EINVAL;
@@ -396,17 +413,7 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
        fl->fl_ops = NULL;
        fl->fl_lmops = NULL;
 
-       switch (l->l_type) {
-       case F_RDLCK:
-       case F_WRLCK:
-       case F_UNLCK:
-               fl->fl_type = l->l_type;
-               break;
-       default:
-               return -EINVAL;
-       }
-
-       return (0);
+       return assign_type(fl, l->l_type);
 }
 #endif
 
@@ -416,31 +423,15 @@ static void lease_break_callback(struct file_lock *fl)
        kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
 }
 
-static void lease_release_private_callback(struct file_lock *fl)
-{
-       if (!fl->fl_file)
-               return;
-
-       f_delown(fl->fl_file);
-       fl->fl_file->f_owner.signum = 0;
-}
-
-static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
-{
-       return fl->fl_file == try->fl_file;
-}
-
-static struct lock_manager_operations lease_manager_ops = {
-       .fl_break = lease_break_callback,
-       .fl_release_private = lease_release_private_callback,
-       .fl_mylease = lease_mylease_callback,
-       .fl_change = lease_modify,
+static const struct lock_manager_operations lease_manager_ops = {
+       .lm_break = lease_break_callback,
+       .lm_change = lease_modify,
 };
 
 /*
  * Initialize a lease, use the default lock manager operations
  */
-static int lease_init(struct file *filp, int type, struct file_lock *fl)
+static int lease_init(struct file *filp, long type, struct file_lock *fl)
  {
        if (assign_type(fl, type) != 0)
                return -EINVAL;
@@ -458,7 +449,7 @@ static int lease_init(struct file *filp, int type, struct file_lock *fl)
 }
 
 /* Allocate a file_lock initialised to this type of lease */
-static struct file_lock *lease_alloc(struct file *filp, int type)
+static struct file_lock *lease_alloc(struct file *filp, long type)
 {
        struct file_lock *fl = locks_alloc_lock();
        int error = -ENOMEM;
@@ -487,9 +478,9 @@ static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
  */
 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 {
-       if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
+       if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
                return fl2->fl_lmops == fl1->fl_lmops &&
-                       fl1->fl_lmops->fl_compare_owner(fl1, fl2);
+                       fl1->fl_lmops->lm_compare_owner(fl1, fl2);
        return fl1->fl_owner == fl2->fl_owner;
 }
 
@@ -505,12 +496,13 @@ static void __locks_delete_block(struct file_lock *waiter)
 
 /*
  */
-static void locks_delete_block(struct file_lock *waiter)
+void locks_delete_block(struct file_lock *waiter)
 {
-       lock_kernel();
+       lock_flocks();
        __locks_delete_block(waiter);
-       unlock_kernel();
+       unlock_flocks();
 }
+EXPORT_SYMBOL(locks_delete_block);
 
 /* Insert waiter into blocker's block list.
  * We use a circular list so that processes can be easily woken up in
@@ -534,11 +526,13 @@ static void locks_insert_block(struct file_lock *blocker,
 static void locks_wake_up_blocks(struct file_lock *blocker)
 {
        while (!list_empty(&blocker->fl_block)) {
-               struct file_lock *waiter = list_entry(blocker->fl_block.next,
+               struct file_lock *waiter;
+
+               waiter = list_first_entry(&blocker->fl_block,
                                struct file_lock, fl_block);
                __locks_delete_block(waiter);
-               if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
-                       waiter->fl_lmops->fl_notify(waiter);
+               if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
+                       waiter->fl_lmops->lm_notify(waiter);
                else
                        wake_up(&waiter->fl_wait);
        }
@@ -551,12 +545,11 @@ static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
 {
        list_add(&fl->fl_link, &file_lock_list);
 
+       fl->fl_nspid = get_pid(task_tgid(current));
+
        /* insert into file's list */
        fl->fl_next = *pos;
        *pos = fl;
-
-       if (fl->fl_ops && fl->fl_ops->fl_insert)
-               fl->fl_ops->fl_insert(fl);
 }
 
 /*
@@ -573,15 +566,11 @@ static void locks_delete_lock(struct file_lock **thisfl_p)
        fl->fl_next = NULL;
        list_del_init(&fl->fl_link);
 
-       fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
-       if (fl->fl_fasync != NULL) {
-               printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
-               fl->fl_fasync = NULL;
+       if (fl->fl_nspid) {
+               put_pid(fl->fl_nspid);
+               fl->fl_nspid = NULL;
        }
 
-       if (fl->fl_ops && fl->fl_ops->fl_remove)
-               fl->fl_ops->fl_remove(fl);
-
        locks_wake_up_blocks(fl);
        locks_free_lock(fl);
 }
@@ -632,91 +621,84 @@ static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *s
        return (locks_conflict(caller_fl, sys_fl));
 }
 
-static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
-{
-       int result = 0;
-       DECLARE_WAITQUEUE(wait, current);
-
-       __set_current_state(TASK_INTERRUPTIBLE);
-       add_wait_queue(fl_wait, &wait);
-       if (timeout == 0)
-               schedule();
-       else
-               result = schedule_timeout(timeout);
-       if (signal_pending(current))
-               result = -ERESTARTSYS;
-       remove_wait_queue(fl_wait, &wait);
-       __set_current_state(TASK_RUNNING);
-       return result;
-}
-
-static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
-{
-       int result;
-       locks_insert_block(blocker, waiter);
-       result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
-       __locks_delete_block(waiter);
-       return result;
-}
-
 void
 posix_test_lock(struct file *filp, struct file_lock *fl)
 {
        struct file_lock *cfl;
 
-       lock_kernel();
-       for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
+       lock_flocks();
+       for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
                if (!IS_POSIX(cfl))
                        continue;
-               if (posix_locks_conflict(cfl, fl))
+               if (posix_locks_conflict(fl, cfl))
                        break;
        }
-       if (cfl)
+       if (cfl) {
                __locks_copy_lock(fl, cfl);
-       else
+               if (cfl->fl_nspid)
+                       fl->fl_pid = pid_vnr(cfl->fl_nspid);
+       } else
                fl->fl_type = F_UNLCK;
-       unlock_kernel();
+       unlock_flocks();
        return;
 }
-
 EXPORT_SYMBOL(posix_test_lock);
 
-/* This function tests for deadlock condition before putting a process to
- * sleep. The detection scheme is no longer recursive. Recursive was neat,
- * but dangerous - we risked stack corruption if the lock data was bad, or
- * if the recursion was too deep for any other reason.
+/*
+ * Deadlock detection:
  *
- * We rely on the fact that a task can only be on one lock's wait queue
- * at a time. When we find blocked_task on a wait queue we can re-search
- * with blocked_task equal to that queue's owner, until either blocked_task
- * isn't found, or blocked_task is found on a queue owned by my_task.
+ * We attempt to detect deadlocks that are due purely to posix file
+ * locks.
  *
- * Note: the above assumption may not be true when handling lock requests
- * from a broken NFS client. But broken NFS clients have a lot more to
- * worry about than proper deadlock detection anyway... --okir
+ * We assume that a task can be waiting for at most one lock at a time.
+ * So for any acquired lock, the process holding that lock may be
+ * waiting on at most one other lock.  That lock in turns may be held by
+ * someone waiting for at most one other lock.  Given a requested lock
+ * caller_fl which is about to wait for a conflicting lock block_fl, we
+ * follow this chain of waiters to ensure we are not about to create a
+ * cycle.
+ *
+ * Since we do this before we ever put a process to sleep on a lock, we
+ * are ensured that there is never a cycle; that is what guarantees that
+ * the while() loop in posix_locks_deadlock() eventually completes.
+ *
+ * Note: the above assumption may not be true when handling lock
+ * requests from a broken NFS client. It may also fail in the presence
+ * of tasks (such as posix threads) sharing the same open file table.
+ *
+ * To handle those cases, we just bail out after a few iterations.
  */
+
+#define MAX_DEADLK_ITERATIONS 10
+
+/* Find a lock that the owner of the given block_fl is blocking on. */
+static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
+{
+       struct file_lock *fl;
+
+       list_for_each_entry(fl, &blocked_list, fl_link) {
+               if (posix_same_owner(fl, block_fl))
+                       return fl->fl_next;
+       }
+       return NULL;
+}
+
 static int posix_locks_deadlock(struct file_lock *caller_fl,
                                struct file_lock *block_fl)
 {
-       struct list_head *tmp;
+       int i = 0;
 
-next_task:
-       if (posix_same_owner(caller_fl, block_fl))
-               return 1;
-       list_for_each(tmp, &blocked_list) {
-               struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
-               if (posix_same_owner(fl, block_fl)) {
-                       fl = fl->fl_next;
-                       block_fl = fl;
-                       goto next_task;
-               }
+       while ((block_fl = what_owner_is_waiting_for(block_fl))) {
+               if (i++ > MAX_DEADLK_ITERATIONS)
+                       return 0;
+               if (posix_same_owner(caller_fl, block_fl))
+                       return 1;
        }
        return 0;
 }
 
 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
- * at the head of the list, but that's secret knowledge known only to
- * flock_lock_file and posix_lock_file.
+ * after any leases, but before any posix locks.
  *
  * Note that if called with an FL_EXISTS argument, the caller may determine
  * whether or not a lock was successfully freed by testing the return
@@ -726,13 +708,20 @@ static int flock_lock_file(struct file *filp, struct file_lock *request)
 {
        struct file_lock *new_fl = NULL;
        struct file_lock **before;
-       struct inode * inode = filp->f_path.dentry->d_inode;
+       struct inode * inode = file_inode(filp);
        int error = 0;
        int found = 0;
 
-       lock_kernel();
+       if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
+               new_fl = locks_alloc_lock();
+               if (!new_fl)
+                       return -ENOMEM;
+       }
+
+       lock_flocks();
        if (request->fl_flags & FL_ACCESS)
                goto find_conflict;
+
        for_each_lock(inode, before) {
                struct file_lock *fl = *before;
                if (IS_POSIX(fl))
@@ -754,16 +743,15 @@ static int flock_lock_file(struct file *filp, struct file_lock *request)
                goto out;
        }
 
-       error = -ENOMEM;
-       new_fl = locks_alloc_lock();
-       if (new_fl == NULL)
-               goto out;
        /*
         * If a higher-priority process was blocked on the old file lock,
         * give it the opportunity to lock the file.
         */
-       if (found)
+       if (found) {
+               unlock_flocks();
                cond_resched();
+               lock_flocks();
+       }
 
 find_conflict:
        for_each_lock(inode, before) {
@@ -775,19 +763,21 @@ find_conflict:
                if (!flock_locks_conflict(request, fl))
                        continue;
                error = -EAGAIN;
-               if (request->fl_flags & FL_SLEEP)
-                       locks_insert_block(fl, request);
+               if (!(request->fl_flags & FL_SLEEP))
+                       goto out;
+               error = FILE_LOCK_DEFERRED;
+               locks_insert_block(fl, request);
                goto out;
        }
        if (request->fl_flags & FL_ACCESS)
                goto out;
        locks_copy_lock(new_fl, request);
-       locks_insert_lock(&inode->i_flock, new_fl);
+       locks_insert_lock(before, new_fl);
        new_fl = NULL;
        error = 0;
 
 out:
-       unlock_kernel();
+       unlock_flocks();
        if (new_fl)
                locks_free_lock(new_fl);
        return error;
@@ -816,23 +806,23 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                new_fl2 = locks_alloc_lock();
        }
 
-       lock_kernel();
+       lock_flocks();
        if (request->fl_type != F_UNLCK) {
                for_each_lock(inode, before) {
-                       struct file_lock *fl = *before;
+                       fl = *before;
                        if (!IS_POSIX(fl))
                                continue;
                        if (!posix_locks_conflict(request, fl))
                                continue;
                        if (conflock)
-                               locks_copy_lock(conflock, fl);
+                               __locks_copy_lock(conflock, fl);
                        error = -EAGAIN;
                        if (!(request->fl_flags & FL_SLEEP))
                                goto out;
                        error = -EDEADLK;
                        if (posix_locks_deadlock(request, fl))
                                goto out;
-                       error = -EAGAIN;
+                       error = FILE_LOCK_DEFERRED;
                        locks_insert_block(fl, request);
                        goto out;
                }
@@ -984,7 +974,7 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                locks_wake_up_blocks(left);
        }
  out:
-       unlock_kernel();
+       unlock_flocks();
        /*
         * Free any unused locks.
         */
@@ -1012,7 +1002,7 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
 int posix_lock_file(struct file *filp, struct file_lock *fl,
                        struct file_lock *conflock)
 {
-       return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
+       return __posix_lock_file(file_inode(filp), fl, conflock);
 }
 EXPORT_SYMBOL(posix_lock_file);
 
@@ -1031,7 +1021,7 @@ int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
        might_sleep ();
        for (;;) {
                error = posix_lock_file(filp, fl, NULL);
-               if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
+               if (error != FILE_LOCK_DEFERRED)
                        break;
                error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
                if (!error)
@@ -1059,14 +1049,14 @@ int locks_mandatory_locked(struct inode *inode)
        /*
         * Search the lock list for this inode for any POSIX locks.
         */
-       lock_kernel();
+       lock_flocks();
        for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
                if (!IS_POSIX(fl))
                        continue;
                if (fl->fl_owner != owner)
                        break;
        }
-       unlock_kernel();
+       unlock_flocks();
        return fl ? -EAGAIN : 0;
 }
 
@@ -1103,9 +1093,7 @@ int locks_mandatory_area(int read_write, struct inode *inode,
 
        for (;;) {
                error = __posix_lock_file(inode, &fl, NULL);
-               if (error != -EAGAIN)
-                       break;
-               if (!(fl.fl_flags & FL_SLEEP))
+               if (error != FILE_LOCK_DEFERRED)
                        break;
                error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
                if (!error) {
@@ -1113,7 +1101,7 @@ int locks_mandatory_area(int read_write, struct inode *inode,
                         * If we've been sleeping someone might have
                         * changed the permissions behind our back.
                         */
-                       if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
+                       if (__mandatory_lock(inode))
                                continue;
                }
 
@@ -1126,6 +1114,17 @@ int locks_mandatory_area(int read_write, struct inode *inode,
 
 EXPORT_SYMBOL(locks_mandatory_area);
 
+static void lease_clear_pending(struct file_lock *fl, int arg)
+{
+       switch (arg) {
+       case F_UNLCK:
+               fl->fl_flags &= ~FL_UNLOCK_PENDING;
+               /* fall through: */
+       case F_RDLCK:
+               fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
+       }
+}
+
 /* We already had a lease on this file; just change its type */
 int lease_modify(struct file_lock **before, int arg)
 {
@@ -1134,27 +1133,44 @@ int lease_modify(struct file_lock **before, int arg)
 
        if (error)
                return error;
+       lease_clear_pending(fl, arg);
        locks_wake_up_blocks(fl);
-       if (arg == F_UNLCK)
+       if (arg == F_UNLCK) {
+               struct file *filp = fl->fl_file;
+
+               f_delown(filp);
+               filp->f_owner.signum = 0;
+               fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
+               if (fl->fl_fasync != NULL) {
+                       printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
+                       fl->fl_fasync = NULL;
+               }
                locks_delete_lock(before);
+       }
        return 0;
 }
 
 EXPORT_SYMBOL(lease_modify);
 
+static bool past_time(unsigned long then)
+{
+       if (!then)
+               /* 0 is a special value meaning "this never expires": */
+               return false;
+       return time_after(jiffies, then);
+}
+
 static void time_out_leases(struct inode *inode)
 {
        struct file_lock **before;
        struct file_lock *fl;
 
        before = &inode->i_flock;
-       while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
-               if ((fl->fl_break_time == 0)
-                               || time_before(jiffies, fl->fl_break_time)) {
-                       before = &fl->fl_next;
-                       continue;
-               }
-               lease_modify(before, fl->fl_type & ~F_INPROGRESS);
+       while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
+               if (past_time(fl->fl_downgrade_time))
+                       lease_modify(before, F_RDLCK);
+               if (past_time(fl->fl_break_time))
+                       lease_modify(before, F_UNLCK);
                if (fl == *before)      /* lease_modify may have freed fl */
                        before = &fl->fl_next;
        }
@@ -1172,15 +1188,18 @@ static void time_out_leases(struct inode *inode)
  */
 int __break_lease(struct inode *inode, unsigned int mode)
 {
-       int error = 0, future;
+       int error = 0;
        struct file_lock *new_fl, *flock;
        struct file_lock *fl;
        unsigned long break_time;
        int i_have_this_lease = 0;
+       int want_write = (mode & O_ACCMODE) != O_RDONLY;
 
-       new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
+       new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
+       if (IS_ERR(new_fl))
+               return PTR_ERR(new_fl);
 
-       lock_kernel();
+       lock_flocks();
 
        time_out_leases(inode);
 
@@ -1188,30 +1207,13 @@ int __break_lease(struct inode *inode, unsigned int mode)
        if ((flock == NULL) || !IS_LEASE(flock))
                goto out;
 
+       if (!locks_conflict(flock, new_fl))
+               goto out;
+
        for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
                if (fl->fl_owner == current->files)
                        i_have_this_lease = 1;
 
-       if (mode & FMODE_WRITE) {
-               /* If we want write access, we have to revoke any lease. */
-               future = F_UNLCK | F_INPROGRESS;
-       } else if (flock->fl_type & F_INPROGRESS) {
-               /* If the lease is already being broken, we just leave it */
-               future = flock->fl_type;
-       } else if (flock->fl_type & F_WRLCK) {
-               /* Downgrade the exclusive lease to a read-only lease. */
-               future = F_RDLCK | F_INPROGRESS;
-       } else {
-               /* the existing lease was read-only, so we can read too. */
-               goto out;
-       }
-
-       if (IS_ERR(new_fl) && !i_have_this_lease
-                       && ((mode & O_NONBLOCK) == 0)) {
-               error = PTR_ERR(new_fl);
-               goto out;
-       }
-
        break_time = 0;
        if (lease_break_time > 0) {
                break_time = jiffies + lease_break_time * HZ;
@@ -1220,12 +1222,18 @@ int __break_lease(struct inode *inode, unsigned int mode)
        }
 
        for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
-               if (fl->fl_type != future) {
-                       fl->fl_type = future;
+               if (want_write) {
+                       if (fl->fl_flags & FL_UNLOCK_PENDING)
+                               continue;
+                       fl->fl_flags |= FL_UNLOCK_PENDING;
                        fl->fl_break_time = break_time;
-                       /* lease must have lmops break callback */
-                       fl->fl_lmops->fl_break(fl);
+               } else {
+                       if (lease_breaking(flock))
+                               continue;
+                       fl->fl_flags |= FL_DOWNGRADE_PENDING;
+                       fl->fl_downgrade_time = break_time;
                }
+               fl->fl_lmops->lm_break(fl);
        }
 
        if (i_have_this_lease || (mode & O_NONBLOCK)) {
@@ -1240,41 +1248,48 @@ restart:
                if (break_time == 0)
                        break_time++;
        }
-       error = locks_block_on_timeout(flock, new_fl, break_time);
+       locks_insert_block(flock, new_fl);
+       unlock_flocks();
+       error = wait_event_interruptible_timeout(new_fl->fl_wait,
+                                               !new_fl->fl_next, break_time);
+       lock_flocks();
+       __locks_delete_block(new_fl);
        if (error >= 0) {
                if (error == 0)
                        time_out_leases(inode);
-               /* Wait for the next lease that has not been broken yet */
+               /*
+                * Wait for the next conflicting lease that has not been
+                * broken yet
+                */
                for (flock = inode->i_flock; flock && IS_LEASE(flock);
                                flock = flock->fl_next) {
-                       if (flock->fl_type & F_INPROGRESS)
+                       if (locks_conflict(new_fl, flock))
                                goto restart;
                }
                error = 0;
        }
 
 out:
-       unlock_kernel();
-       if (!IS_ERR(new_fl))
-               locks_free_lock(new_fl);
+       unlock_flocks();
+       locks_free_lock(new_fl);
        return error;
 }
 
 EXPORT_SYMBOL(__break_lease);
 
 /**
- *     lease_get_mtime
+ *     lease_get_mtime - get the last modified time of an inode
  *     @inode: the inode
  *      @time:  pointer to a timespec which will contain the last modified time
  *
  * This is to force NFS clients to flush their caches for files with
  * exclusive leases.  The justification is that if someone has an
- * exclusive lease, then they could be modifiying it.
+ * exclusive lease, then they could be modifying it.
  */
 void lease_get_mtime(struct inode *inode, struct timespec *time)
 {
        struct file_lock *flock = inode->i_flock;
-       if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
+       if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
                *time = current_fs_time(inode->i_sb);
        else
                *time = inode->i_mtime;
@@ -1310,48 +1325,25 @@ int fcntl_getlease(struct file *filp)
        struct file_lock *fl;
        int type = F_UNLCK;
 
-       lock_kernel();
-       time_out_leases(filp->f_path.dentry->d_inode);
-       for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
+       lock_flocks();
+       time_out_leases(file_inode(filp));
+       for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
                        fl = fl->fl_next) {
                if (fl->fl_file == filp) {
-                       type = fl->fl_type & ~F_INPROGRESS;
+                       type = target_leasetype(fl);
                        break;
                }
        }
-       unlock_kernel();
+       unlock_flocks();
        return type;
 }
 
-/**
- *     setlease        -       sets a lease on an open file
- *     @filp: file pointer
- *     @arg: type of lease to obtain
- *     @flp: input - file_lock to use, output - file_lock inserted
- *
- *     The (input) flp->fl_lmops->fl_break function is required
- *     by break_lease().
- *
- *     Called with kernel lock held.
- */
-int setlease(struct file *filp, long arg, struct file_lock **flp)
+int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
 {
        struct file_lock *fl, **before, **my_before = NULL, *lease;
        struct dentry *dentry = filp->f_path.dentry;
        struct inode *inode = dentry->d_inode;
-       int error, rdlease_count = 0, wrlease_count = 0;
-
-       if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
-               return -EACCES;
-       if (!S_ISREG(inode->i_mode))
-               return -EINVAL;
-       error = security_file_lock(filp, arg);
-       if (error)
-               return error;
-
-       time_out_leases(inode);
-
-       BUG_ON(!(*flp)->fl_lmops->fl_break);
+       int error;
 
        lease = *flp;
 
@@ -1359,7 +1351,7 @@ int setlease(struct file *filp, long arg, struct file_lock **flp)
        if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
                goto out;
        if ((arg == F_WRLCK)
-           && ((atomic_read(&dentry->d_count) > 1)
+           && ((dentry->d_count > 1)
                || (atomic_read(&inode->i_count) > 1)))
                goto out;
 
@@ -1371,64 +1363,119 @@ int setlease(struct file *filp, long arg, struct file_lock **flp)
         * then the file is not open by anyone (including us)
         * except for this filp.
         */
+       error = -EAGAIN;
        for (before = &inode->i_flock;
                        ((fl = *before) != NULL) && IS_LEASE(fl);
                        before = &fl->fl_next) {
-               if (lease->fl_lmops->fl_mylease(fl, lease))
+               if (fl->fl_file == filp) {
                        my_before = before;
-               else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
-                       /*
-                        * Someone is in the process of opening this
-                        * file for writing so we may not take an
-                        * exclusive lease on it.
-                        */
-                       wrlease_count++;
-               else
-                       rdlease_count++;
+                       continue;
+               }
+               /*
+                * No exclusive leases if someone else has a lease on
+                * this file:
+                */
+               if (arg == F_WRLCK)
+                       goto out;
+               /*
+                * Modifying our existing lease is OK, but no getting a
+                * new lease if someone else is opening for write:
+                */
+               if (fl->fl_flags & FL_UNLOCK_PENDING)
+                       goto out;
        }
 
-       if ((arg == F_RDLCK && (wrlease_count > 0)) ||
-           (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
-               goto out;
-
        if (my_before != NULL) {
-               *flp = *my_before;
-               error = lease->fl_lmops->fl_change(my_before, arg);
+               error = lease->fl_lmops->lm_change(my_before, arg);
+               if (!error)
+                       *flp = *my_before;
                goto out;
        }
 
-       error = 0;
-       if (arg == F_UNLCK)
-               goto out;
-
        error = -EINVAL;
        if (!leases_enable)
                goto out;
 
-       error = -ENOMEM;
-       fl = locks_alloc_lock();
-       if (fl == NULL)
-               goto out;
-
-       locks_copy_lock(fl, lease);
-
-       locks_insert_lock(before, fl);
+       locks_insert_lock(before, lease);
+       return 0;
 
-       *flp = fl;
-       error = 0;
 out:
        return error;
 }
-EXPORT_SYMBOL(setlease);
 
- /**
+int generic_delete_lease(struct file *filp, struct file_lock **flp)
+{
+       struct file_lock *fl, **before;
+       struct dentry *dentry = filp->f_path.dentry;
+       struct inode *inode = dentry->d_inode;
+
+       for (before = &inode->i_flock;
+                       ((fl = *before) != NULL) && IS_LEASE(fl);
+                       before = &fl->fl_next) {
+               if (fl->fl_file != filp)
+                       continue;
+               return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
+       }
+       return -EAGAIN;
+}
+
+/**
+ *     generic_setlease        -       sets a lease on an open file
+ *     @filp: file pointer
+ *     @arg: type of lease to obtain
+ *     @flp: input - file_lock to use, output - file_lock inserted
+ *
+ *     The (input) flp->fl_lmops->lm_break function is required
+ *     by break_lease().
+ *
+ *     Called with file_lock_lock held.
+ */
+int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
+{
+       struct dentry *dentry = filp->f_path.dentry;
+       struct inode *inode = dentry->d_inode;
+       int error;
+
+       if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
+               return -EACCES;
+       if (!S_ISREG(inode->i_mode))
+               return -EINVAL;
+       error = security_file_lock(filp, arg);
+       if (error)
+               return error;
+
+       time_out_leases(inode);
+
+       BUG_ON(!(*flp)->fl_lmops->lm_break);
+
+       switch (arg) {
+       case F_UNLCK:
+               return generic_delete_lease(filp, flp);
+       case F_RDLCK:
+       case F_WRLCK:
+               return generic_add_lease(filp, arg, flp);
+       default:
+               return -EINVAL;
+       }
+}
+EXPORT_SYMBOL(generic_setlease);
+
+static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
+{
+       if (filp->f_op && filp->f_op->setlease)
+               return filp->f_op->setlease(filp, arg, lease);
+       else
+               return generic_setlease(filp, arg, lease);
+}
+
+/**
  *     vfs_setlease        -       sets a lease on an open file
  *     @filp: file pointer
  *     @arg: type of lease to obtain
  *     @lease: file_lock to use
  *
  *     Call this to establish a lease on the file.
- *     The (*lease)->fl_lmops->fl_break operation must be set; if not,
+ *     The (*lease)->fl_lmops->lm_break operation must be set; if not,
  *     break_lease will oops!
  *
  *     This will call the filesystem's setlease file method, if
@@ -1440,7 +1487,7 @@ EXPORT_SYMBOL(setlease);
  *     leases held by processes on this node.
  *
  *     There is also no break_lease method; filesystems that
- *     handle their own leases shoud break leases themselves from the
+ *     handle their own leases should break leases themselves from the
  *     filesystem's open, create, and (on truncate) setattr methods.
  *
  *     Warning: the only current setlease methods exist only to disable
@@ -1452,17 +1499,67 @@ int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
 {
        int error;
 
-       lock_kernel();
-       if (filp->f_op && filp->f_op->setlease)
-               error = filp->f_op->setlease(filp, arg, lease);
-       else
-               error = setlease(filp, arg, lease);
-       unlock_kernel();
+       lock_flocks();
+       error = __vfs_setlease(filp, arg, lease);
+       unlock_flocks();
 
        return error;
 }
 EXPORT_SYMBOL_GPL(vfs_setlease);
 
+static int do_fcntl_delete_lease(struct file *filp)
+{
+       struct file_lock fl, *flp = &fl;
+
+       lease_init(filp, F_UNLCK, flp);
+
+       return vfs_setlease(filp, F_UNLCK, &flp);
+}
+
+static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
+{
+       struct file_lock *fl, *ret;
+       struct fasync_struct *new;
+       int error;
+
+       fl = lease_alloc(filp, arg);
+       if (IS_ERR(fl))
+               return PTR_ERR(fl);
+
+       new = fasync_alloc();
+       if (!new) {
+               locks_free_lock(fl);
+               return -ENOMEM;
+       }
+       ret = fl;
+       lock_flocks();
+       error = __vfs_setlease(filp, arg, &ret);
+       if (error) {
+               unlock_flocks();
+               locks_free_lock(fl);
+               goto out_free_fasync;
+       }
+       if (ret != fl)
+               locks_free_lock(fl);
+
+       /*
+        * fasync_insert_entry() returns the old entry if any.
+        * If there was no old entry, then it used 'new' and
+        * inserted it into the fasync list. Clear new so that
+        * we don't release it here.
+        */
+       if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
+               new = NULL;
+
+       error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
+       unlock_flocks();
+
+out_free_fasync:
+       if (new)
+               fasync_free(new);
+       return error;
+}
+
 /**
  *     fcntl_setlease  -       sets a lease on an open file
  *     @fd: open file descriptor
@@ -1475,35 +1572,9 @@ EXPORT_SYMBOL_GPL(vfs_setlease);
  */
 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
 {
-       struct file_lock fl, *flp = &fl;
-       struct dentry *dentry = filp->f_path.dentry;
-       struct inode *inode = dentry->d_inode;
-       int error;
-
-       locks_init_lock(&fl);
-       error = lease_init(filp, arg, &fl);
-       if (error)
-               return error;
-
-       lock_kernel();
-
-       error = vfs_setlease(filp, arg, &flp);
-       if (error || arg == F_UNLCK)
-               goto out_unlock;
-
-       error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
-       if (error < 0) {
-               /* remove lease just inserted by setlease */
-               flp->fl_type = F_UNLCK | F_INPROGRESS;
-               flp->fl_break_time = jiffies - 10;
-               time_out_leases(inode);
-               goto out_unlock;
-       }
-
-       error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
-out_unlock:
-       unlock_kernel();
-       return error;
+       if (arg == F_UNLCK)
+               return do_fcntl_delete_lease(filp);
+       return do_fcntl_add_lease(fd, filp, arg);
 }
 
 /**
@@ -1519,7 +1590,7 @@ int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
        might_sleep();
        for (;;) {
                error = flock_lock_file(filp, fl);
-               if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
+               if (error != FILE_LOCK_DEFERRED)
                        break;
                error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
                if (!error)
@@ -1552,47 +1623,47 @@ EXPORT_SYMBOL(flock_lock_file_wait);
  *     %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
  *     processes read and write access respectively.
  */
-asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
+SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
 {
-       struct file *filp;
+       struct fd f = fdget(fd);
        struct file_lock *lock;
        int can_sleep, unlock;
        int error;
 
        error = -EBADF;
-       filp = fget(fd);
-       if (!filp)
+       if (!f.file)
                goto out;
 
        can_sleep = !(cmd & LOCK_NB);
        cmd &= ~LOCK_NB;
        unlock = (cmd == LOCK_UN);
 
-       if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
+       if (!unlock && !(cmd & LOCK_MAND) &&
+           !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
                goto out_putf;
 
-       error = flock_make_lock(filp, &lock, cmd);
+       error = flock_make_lock(f.file, &lock, cmd);
        if (error)
                goto out_putf;
        if (can_sleep)
                lock->fl_flags |= FL_SLEEP;
 
-       error = security_file_lock(filp, cmd);
+       error = security_file_lock(f.file, lock->fl_type);
        if (error)
                goto out_free;
 
-       if (filp->f_op && filp->f_op->flock)
-               error = filp->f_op->flock(filp,
+       if (f.file->f_op && f.file->f_op->flock)
+               error = f.file->f_op->flock(f.file,
                                          (can_sleep) ? F_SETLKW : F_SETLK,
                                          lock);
        else
-               error = flock_lock_file_wait(filp, lock);
+               error = flock_lock_file_wait(f.file, lock);
 
  out_free:
        locks_free_lock(lock);
 
  out_putf:
-       fput(filp);
+       fdput(f);
  out:
        return error;
 }
@@ -1600,8 +1671,7 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
 /**
  * vfs_test_lock - test file byte range lock
  * @filp: The file to test lock for
- * @fl: The lock to test
- * @conf: Place to return a copy of the conflicting lock, if found
+ * @fl: The lock to test; also used to hold result
  *
  * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
  * setting conf->fl_type to something other than F_UNLCK.
@@ -1702,20 +1772,20 @@ out:
  * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
  * locks, the ->lock() interface may return asynchronously, before the lock has
  * been granted or denied by the underlying filesystem, if (and only if)
- * fl_grant is set. Callers expecting ->lock() to return asynchronously
+ * lm_grant is set. Callers expecting ->lock() to return asynchronously
  * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
  * the request is for a blocking lock. When ->lock() does return asynchronously,
- * it must return -EINPROGRESS, and call ->fl_grant() when the lock
+ * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
  * request completes.
  * If the request is for non-blocking lock the file system should return
- * -EINPROGRESS then try to get the lock and call the callback routine with
- * the result. If the request timed out the callback routine will return a
+ * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
+ * with the result. If the request timed out the callback routine will return a
  * nonzero return code and the file system should release the lock. The file
  * system is also responsible to keep a corresponding posix lock when it
  * grants a lock so the VFS can find out which locks are locally held and do
  * the correct lock cleanup when required.
  * The underlying filesystem must not drop the kernel lock or call
- * ->fl_grant() before returning to the caller with a -EINPROGRESS
+ * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
  * return code.
  */
 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
@@ -1727,6 +1797,30 @@ int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, str
 }
 EXPORT_SYMBOL_GPL(vfs_lock_file);
 
+static int do_lock_file_wait(struct file *filp, unsigned int cmd,
+                            struct file_lock *fl)
+{
+       int error;
+
+       error = security_file_lock(filp, fl->fl_type);
+       if (error)
+               return error;
+
+       for (;;) {
+               error = vfs_lock_file(filp, cmd, fl, NULL);
+               if (error != FILE_LOCK_DEFERRED)
+                       break;
+               error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
+               if (!error)
+                       continue;
+
+               locks_delete_block(fl);
+               break;
+       }
+
+       return error;
+}
+
 /* Apply the lock described by l to an open file descriptor.
  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
  */
@@ -1736,6 +1830,7 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
        struct file_lock *file_lock = locks_alloc_lock();
        struct flock flock;
        struct inode *inode;
+       struct file *f;
        int error;
 
        if (file_lock == NULL)
@@ -1748,14 +1843,12 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
        if (copy_from_user(&flock, l, sizeof(flock)))
                goto out;
 
-       inode = filp->f_path.dentry->d_inode;
+       inode = file_inode(filp);
 
        /* Don't allow mandatory locks on files that may be memory mapped
         * and shared.
         */
-       if (IS_MANDLOCK(inode) &&
-           (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
-           mapping_writably_mapped(filp->f_mapping)) {
+       if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
                error = -EAGAIN;
                goto out;
        }
@@ -1785,28 +1878,21 @@ again:
                goto out;
        }
 
-       error = security_file_lock(filp, file_lock->fl_type);
-       if (error)
-               goto out;
-
-       for (;;) {
-               error = vfs_lock_file(filp, cmd, file_lock, NULL);
-               if (error != -EAGAIN || cmd == F_SETLK)
-                       break;
-               error = wait_event_interruptible(file_lock->fl_wait,
-                               !file_lock->fl_next);
-               if (!error)
-                       continue;
-
-               locks_delete_block(file_lock);
-               break;
-       }
+       error = do_lock_file_wait(filp, cmd, file_lock);
 
        /*
         * Attempt to detect a close/fcntl race and recover by
         * releasing the lock that was just acquired.
         */
-       if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
+       /*
+        * we need that spin_lock here - it prevents reordering between
+        * update of inode->i_flock and check for it done in close().
+        * rcu_read_lock() wouldn't do.
+        */
+       spin_lock(&current->files->file_lock);
+       f = fcheck(fd);
+       spin_unlock(&current->files->file_lock);
+       if (!error && f != filp && flock.l_type != F_UNLCK) {
                flock.l_type = F_UNLCK;
                goto again;
        }
@@ -1862,6 +1948,7 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
        struct file_lock *file_lock = locks_alloc_lock();
        struct flock64 flock;
        struct inode *inode;
+       struct file *f;
        int error;
 
        if (file_lock == NULL)
@@ -1874,14 +1961,12 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
        if (copy_from_user(&flock, l, sizeof(flock)))
                goto out;
 
-       inode = filp->f_path.dentry->d_inode;
+       inode = file_inode(filp);
 
        /* Don't allow mandatory locks on files that may be memory mapped
         * and shared.
         */
-       if (IS_MANDLOCK(inode) &&
-           (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
-           mapping_writably_mapped(filp->f_mapping)) {
+       if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
                error = -EAGAIN;
                goto out;
        }
@@ -1911,28 +1996,16 @@ again:
                goto out;
        }
 
-       error = security_file_lock(filp, file_lock->fl_type);
-       if (error)
-               goto out;
-
-       for (;;) {
-               error = vfs_lock_file(filp, cmd, file_lock, NULL);
-               if (error != -EAGAIN || cmd == F_SETLK64)
-                       break;
-               error = wait_event_interruptible(file_lock->fl_wait,
-                               !file_lock->fl_next);
-               if (!error)
-                       continue;
-
-               locks_delete_block(file_lock);
-               break;
-       }
+       error = do_lock_file_wait(filp, cmd, file_lock);
 
        /*
         * Attempt to detect a close/fcntl race and recover by
         * releasing the lock that was just acquired.
         */
-       if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
+       spin_lock(&current->files->file_lock);
+       f = fcheck(fd);
+       spin_unlock(&current->files->file_lock);
+       if (!error && f != filp && flock.l_type != F_UNLCK) {
                flock.l_type = F_UNLCK;
                goto again;
        }
@@ -1957,7 +2030,7 @@ void locks_remove_posix(struct file *filp, fl_owner_t owner)
         * posix_lock_file().  Another process could be setting a lock on this
         * file at the same time, but we wouldn't remove that lock anyway.
         */
-       if (!filp->f_path.dentry->d_inode->i_flock)
+       if (!file_inode(filp)->i_flock)
                return;
 
        lock.fl_type = F_UNLCK;
@@ -1983,7 +2056,7 @@ EXPORT_SYMBOL(locks_remove_posix);
  */
 void locks_remove_flock(struct file *filp)
 {
-       struct inode * inode = filp->f_path.dentry->d_inode;
+       struct inode * inode = file_inode(filp);
        struct file_lock *fl;
        struct file_lock **before;
 
@@ -2003,7 +2076,7 @@ void locks_remove_flock(struct file *filp)
                        fl.fl_ops->fl_release_private(&fl);
        }
 
-       lock_kernel();
+       lock_flocks();
        before = &inode->i_flock;
 
        while ((fl = *before) != NULL) {
@@ -2021,7 +2094,7 @@ void locks_remove_flock(struct file *filp)
                }
                before = &fl->fl_next;
        }
-       unlock_kernel();
+       unlock_flocks();
 }
 
 /**
@@ -2036,12 +2109,12 @@ posix_unblock_lock(struct file *filp, struct file_lock *waiter)
 {
        int status = 0;
 
-       lock_kernel();
+       lock_flocks();
        if (waiter->fl_next)
                __locks_delete_block(waiter);
        else
                status = -ENOENT;
-       unlock_kernel();
+       unlock_flocks();
        return status;
 }
 
@@ -2063,137 +2136,142 @@ int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
 
 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
 
-static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
+#ifdef CONFIG_PROC_FS
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+
+static void lock_get_status(struct seq_file *f, struct file_lock *fl,
+                           loff_t id, char *pfx)
 {
        struct inode *inode = NULL;
+       unsigned int fl_pid;
+
+       if (fl->fl_nspid)
+               fl_pid = pid_vnr(fl->fl_nspid);
+       else
+               fl_pid = fl->fl_pid;
 
        if (fl->fl_file != NULL)
-               inode = fl->fl_file->f_path.dentry->d_inode;
+               inode = file_inode(fl->fl_file);
 
-       out += sprintf(out, "%d:%s ", id, pfx);
+       seq_printf(f, "%lld:%s ", id, pfx);
        if (IS_POSIX(fl)) {
-               out += sprintf(out, "%6s %s ",
+               seq_printf(f, "%6s %s ",
                             (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
                             (inode == NULL) ? "*NOINODE*" :
-                            (IS_MANDLOCK(inode) &&
-                             (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
-                            "MANDATORY" : "ADVISORY ");
+                            mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
        } else if (IS_FLOCK(fl)) {
                if (fl->fl_type & LOCK_MAND) {
-                       out += sprintf(out, "FLOCK  MSNFS     ");
+                       seq_printf(f, "FLOCK  MSNFS     ");
                } else {
-                       out += sprintf(out, "FLOCK  ADVISORY  ");
+                       seq_printf(f, "FLOCK  ADVISORY  ");
                }
        } else if (IS_LEASE(fl)) {
-               out += sprintf(out, "LEASE  ");
-               if (fl->fl_type & F_INPROGRESS)
-                       out += sprintf(out, "BREAKING  ");
+               seq_printf(f, "LEASE  ");
+               if (lease_breaking(fl))
+                       seq_printf(f, "BREAKING  ");
                else if (fl->fl_file)
-                       out += sprintf(out, "ACTIVE    ");
+                       seq_printf(f, "ACTIVE    ");
                else
-                       out += sprintf(out, "BREAKER   ");
+                       seq_printf(f, "BREAKER   ");
        } else {
-               out += sprintf(out, "UNKNOWN UNKNOWN  ");
+               seq_printf(f, "UNKNOWN UNKNOWN  ");
        }
        if (fl->fl_type & LOCK_MAND) {
-               out += sprintf(out, "%s ",
+               seq_printf(f, "%s ",
                               (fl->fl_type & LOCK_READ)
                               ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
                               : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
        } else {
-               out += sprintf(out, "%s ",
-                              (fl->fl_type & F_INPROGRESS)
-                              ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
-                              : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
+               seq_printf(f, "%s ",
+                              (lease_breaking(fl))
+                              ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
+                              : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
        }
        if (inode) {
 #ifdef WE_CAN_BREAK_LSLK_NOW
-               out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
+               seq_printf(f, "%d %s:%ld ", fl_pid,
                                inode->i_sb->s_id, inode->i_ino);
 #else
                /* userspace relies on this representation of dev_t ;-( */
-               out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
+               seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
                                MAJOR(inode->i_sb->s_dev),
                                MINOR(inode->i_sb->s_dev), inode->i_ino);
 #endif
        } else {
-               out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
+               seq_printf(f, "%d <none>:0 ", fl_pid);
        }
        if (IS_POSIX(fl)) {
                if (fl->fl_end == OFFSET_MAX)
-                       out += sprintf(out, "%Ld EOF\n", fl->fl_start);
+                       seq_printf(f, "%Ld EOF\n", fl->fl_start);
                else
-                       out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
-                                       fl->fl_end);
+                       seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
        } else {
-               out += sprintf(out, "0 EOF\n");
+               seq_printf(f, "0 EOF\n");
        }
 }
 
-static void move_lock_status(char **p, off_t* pos, off_t offset)
+static int locks_show(struct seq_file *f, void *v)
 {
-       int len;
-       len = strlen(*p);
-       if(*pos >= offset) {
-               /* the complete line is valid */
-               *p += len;
-               *pos += len;
-               return;
-       }
-       if(*pos+len > offset) {
-               /* use the second part of the line */
-               int i = offset-*pos;
-               memmove(*p,*p+i,len-i);
-               *p += len-i;
-               *pos += len;
-               return;
-       }
-       /* discard the complete line */
-       *pos += len;
+       struct file_lock *fl, *bfl;
+
+       fl = list_entry(v, struct file_lock, fl_link);
+
+       lock_get_status(f, fl, *((loff_t *)f->private), "");
+
+       list_for_each_entry(bfl, &fl->fl_block, fl_block)
+               lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
+
+       return 0;
 }
 
-/**
- *     get_locks_status        -       reports lock usage in /proc/locks
- *     @buffer: address in userspace to write into
- *     @start: ?
- *     @offset: how far we are through the buffer
- *     @length: how much to read
- */
+static void *locks_start(struct seq_file *f, loff_t *pos)
+{
+       loff_t *p = f->private;
+
+       lock_flocks();
+       *p = (*pos + 1);
+       return seq_list_start(&file_lock_list, *pos);
+}
 
-int get_locks_status(char *buffer, char **start, off_t offset, int length)
+static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
 {
-       struct list_head *tmp;
-       char *q = buffer;
-       off_t pos = 0;
-       int i = 0;
+       loff_t *p = f->private;
+       ++*p;
+       return seq_list_next(v, &file_lock_list, pos);
+}
 
-       lock_kernel();
-       list_for_each(tmp, &file_lock_list) {
-               struct list_head *btmp;
-               struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
-               lock_get_status(q, fl, ++i, "");
-               move_lock_status(&q, &pos, offset);
+static void locks_stop(struct seq_file *f, void *v)
+{
+       unlock_flocks();
+}
 
-               if(pos >= offset+length)
-                       goto done;
+static const struct seq_operations locks_seq_operations = {
+       .start  = locks_start,
+       .next   = locks_next,
+       .stop   = locks_stop,
+       .show   = locks_show,
+};
 
-               list_for_each(btmp, &fl->fl_block) {
-                       struct file_lock *bfl = list_entry(btmp,
-                                       struct file_lock, fl_block);
-                       lock_get_status(q, bfl, i, " ->");
-                       move_lock_status(&q, &pos, offset);
+static int locks_open(struct inode *inode, struct file *filp)
+{
+       return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
+}
 
-                       if(pos >= offset+length)
-                               goto done;
-               }
-       }
-done:
-       unlock_kernel();
-       *start = buffer;
-       if(q-buffer < length)
-               return (q-buffer);
-       return length;
+static const struct file_operations proc_locks_operations = {
+       .open           = locks_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release_private,
+};
+
+static int __init proc_locks_init(void)
+{
+       proc_create("locks", 0, NULL, &proc_locks_operations);
+       return 0;
 }
+module_init(proc_locks_init);
+#endif
 
 /**
  *     lock_may_read - checks that the region is free of locks
@@ -2212,7 +2290,7 @@ int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
 {
        struct file_lock *fl;
        int result = 1;
-       lock_kernel();
+       lock_flocks();
        for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
                if (IS_POSIX(fl)) {
                        if (fl->fl_type == F_RDLCK)
@@ -2229,7 +2307,7 @@ int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
                result = 0;
                break;
        }
-       unlock_kernel();
+       unlock_flocks();
        return result;
 }
 
@@ -2252,7 +2330,7 @@ int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
 {
        struct file_lock *fl;
        int result = 1;
-       lock_kernel();
+       lock_flocks();
        for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
                if (IS_POSIX(fl)) {
                        if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
@@ -2267,7 +2345,7 @@ int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
                result = 0;
                break;
        }
-       unlock_kernel();
+       unlock_flocks();
        return result;
 }
 
@@ -2276,8 +2354,8 @@ EXPORT_SYMBOL(lock_may_write);
 static int __init filelock_init(void)
 {
        filelock_cache = kmem_cache_create("file_lock_cache",
-                       sizeof(struct file_lock), 0, SLAB_PANIC,
-                       init_once, NULL);
+                       sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
+
        return 0;
 }