mmc: s3c6410: enable ADMA feature in 6410 sdhci controller
[safe/jmp/linux-2.6] / fs / locks.c
index 4f2d749..ab24d49 100644 (file)
 
 #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/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)
@@ -150,7 +151,7 @@ static struct file_lock *locks_alloc_lock(void)
        return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
 }
 
-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)
@@ -164,6 +165,7 @@ static void locks_release_private(struct file_lock *fl)
        }
 
 }
+EXPORT_SYMBOL_GPL(locks_release_private);
 
 /* Free a lock which is not in use. */
 static void locks_free_lock(struct file_lock *fl)
@@ -185,6 +187,7 @@ void locks_init_lock(struct file_lock *fl)
        fl->fl_fasync = NULL;
        fl->fl_owner = NULL;
        fl->fl_pid = 0;
+       fl->fl_nspid = NULL;
        fl->fl_file = NULL;
        fl->fl_flags = 0;
        fl->fl_type = 0;
@@ -199,7 +202,7 @@ 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)
+static void init_once(void *foo)
 {
        struct file_lock *lock = (struct file_lock *) foo;
 
@@ -223,7 +226,7 @@ static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
 /*
  * 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 +238,7 @@ 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)
 {
@@ -430,7 +434,7 @@ 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 = {
+static const struct lock_manager_operations lease_manager_ops = {
        .fl_break = lease_break_callback,
        .fl_release_private = lease_release_private_callback,
        .fl_mylease = lease_mylease_callback,
@@ -534,7 +538,9 @@ 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)
@@ -551,12 +557,11 @@ 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);
 }
 
 /*
@@ -579,8 +584,10 @@ static void locks_delete_lock(struct file_lock **thisfl_p)
                fl->fl_fasync = NULL;
        }
 
-       if (fl->fl_ops && fl->fl_ops->fl_remove)
-               fl->fl_ops->fl_remove(fl);
+       if (fl->fl_nspid) {
+               put_pid(fl->fl_nspid);
+               fl->fl_nspid = NULL;
+       }
 
        locks_wake_up_blocks(fl);
        locks_free_lock(fl);
@@ -632,33 +639,6 @@ 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)
 {
@@ -668,55 +648,75 @@ posix_test_lock(struct file *filp, struct file_lock *fl)
        for (cfl = filp->f_path.dentry->d_inode->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();
        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
@@ -733,6 +733,15 @@ static int flock_lock_file(struct file *filp, struct file_lock *request)
        lock_kernel();
        if (request->fl_flags & FL_ACCESS)
                goto find_conflict;
+
+       if (request->fl_type != F_UNLCK) {
+               error = -ENOMEM;
+               new_fl = locks_alloc_lock();
+               if (new_fl == NULL)
+                       goto out;
+               error = 0;
+       }
+
        for_each_lock(inode, before) {
                struct file_lock *fl = *before;
                if (IS_POSIX(fl))
@@ -754,10 +763,6 @@ 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.
@@ -775,14 +780,16 @@ 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;
 
@@ -819,20 +826,20 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
        lock_kernel();
        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;
                }
@@ -1031,7 +1038,7 @@ 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)
@@ -1103,9 +1110,7 @@ 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 +1118,7 @@ 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;
                }
 
@@ -1177,8 +1182,9 @@ int __break_lease(struct inode *inode, unsigned int mode)
        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);
 
        lock_kernel();
 
@@ -1192,7 +1198,7 @@ int __break_lease(struct inode *inode, unsigned int mode)
                if (fl->fl_owner == current->files)
                        i_have_this_lease = 1;
 
-       if (mode & FMODE_WRITE) {
+       if (want_write) {
                /* If we want write access, we have to revoke any lease. */
                future = F_UNLCK | F_INPROGRESS;
        } else if (flock->fl_type & F_INPROGRESS) {
@@ -1240,7 +1246,10 @@ restart:
                if (break_time == 0)
                        break_time++;
        }
-       error = locks_block_on_timeout(flock, new_fl, break_time);
+       locks_insert_block(flock, new_fl);
+       error = wait_event_interruptible_timeout(new_fl->fl_wait,
+                                               !new_fl->fl_next, break_time);
+       __locks_delete_block(new_fl);
        if (error >= 0) {
                if (error == 0)
                        time_out_leases(inode);
@@ -1263,13 +1272,13 @@ out:
 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)
 {
@@ -1324,7 +1333,7 @@ int fcntl_getlease(struct file *filp)
 }
 
 /**
- *     setlease        -       sets a lease on an open file
+ *     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
@@ -1334,14 +1343,15 @@ int fcntl_getlease(struct file *filp)
  *
  *     Called with kernel lock held.
  */
-int setlease(struct file *filp, long arg, struct file_lock **flp)
+int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
 {
        struct file_lock *fl, **before, **my_before = NULL, *lease;
+       struct file_lock *new_fl = NULL;
        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))
+       if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
                return -EACCES;
        if (!S_ISREG(inode->i_mode))
                return -EINVAL;
@@ -1355,13 +1365,20 @@ int setlease(struct file *filp, long arg, struct file_lock **flp)
 
        lease = *flp;
 
-       error = -EAGAIN;
-       if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
-               goto out;
-       if ((arg == F_WRLCK)
-           && ((atomic_read(&dentry->d_count) > 1)
-               || (atomic_read(&inode->i_count) > 1)))
-               goto out;
+       if (arg != F_UNLCK) {
+               error = -ENOMEM;
+               new_fl = locks_alloc_lock();
+               if (new_fl == NULL)
+                       goto out;
+
+               error = -EAGAIN;
+               if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
+                       goto out;
+               if ((arg == F_WRLCK)
+                   && ((atomic_read(&dentry->d_count) > 1)
+                       || (atomic_read(&inode->i_count) > 1)))
+                       goto out;
+       }
 
        /*
         * At this point, we know that if there is an exclusive
@@ -1387,6 +1404,7 @@ int setlease(struct file *filp, long arg, struct file_lock **flp)
                        rdlease_count++;
        }
 
+       error = -EAGAIN;
        if ((arg == F_RDLCK && (wrlease_count > 0)) ||
            (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
                goto out;
@@ -1405,21 +1423,18 @@ int setlease(struct file *filp, long arg, struct file_lock **flp)
        if (!leases_enable)
                goto out;
 
-       error = -ENOMEM;
-       fl = locks_alloc_lock();
-       if (fl == NULL)
-               goto out;
+       locks_copy_lock(new_fl, lease);
+       locks_insert_lock(before, new_fl);
 
-       locks_copy_lock(fl, lease);
-
-       locks_insert_lock(before, fl);
+       *flp = new_fl;
+       return 0;
 
-       *flp = fl;
-       error = 0;
 out:
+       if (new_fl != NULL)
+               locks_free_lock(new_fl);
        return error;
 }
-EXPORT_SYMBOL(setlease);
+EXPORT_SYMBOL(generic_setlease);
 
  /**
  *     vfs_setlease        -       sets a lease on an open file
@@ -1440,7 +1455,7 @@ 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
@@ -1456,7 +1471,7 @@ int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
        if (filp->f_op && filp->f_op->setlease)
                error = filp->f_op->setlease(filp, arg, lease);
        else
-               error = setlease(filp, arg, lease);
+               error = generic_setlease(filp, arg, lease);
        unlock_kernel();
 
        return error;
@@ -1476,8 +1491,7 @@ 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;
+       struct inode *inode = filp->f_path.dentry->d_inode;
        int error;
 
        locks_init_lock(&fl);
@@ -1519,7 +1533,7 @@ 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)
@@ -1552,7 +1566,7 @@ 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 file_lock *lock;
@@ -1568,7 +1582,8 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
        cmd &= ~LOCK_NB;
        unlock = (cmd == LOCK_UN);
 
-       if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
+       if (!unlock && !(cmd & LOCK_MAND) &&
+           !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
                goto out_putf;
 
        error = flock_make_lock(filp, &lock, cmd);
@@ -1577,7 +1592,7 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
        if (can_sleep)
                lock->fl_flags |= FL_SLEEP;
 
-       error = security_file_lock(filp, cmd);
+       error = security_file_lock(filp, lock->fl_type);
        if (error)
                goto out_free;
 
@@ -1704,17 +1719,17 @@ out:
  * fl_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 ->fl_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
+ * ->fl_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)
@@ -1726,6 +1741,30 @@ 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().
  */
@@ -1735,6 +1774,7 @@ 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)
@@ -1752,9 +1792,7 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
        /* 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;
        }
@@ -1784,28 +1822,21 @@ 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;
        }
@@ -1861,6 +1892,7 @@ 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)
@@ -1878,9 +1910,7 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
        /* 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;
        }
@@ -1910,28 +1940,16 @@ 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;
        }
@@ -2062,138 +2080,140 @@ 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,
+                                                       int 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;
 
-       out += sprintf(out, "%d:%s ", id, pfx);
+       seq_printf(f, "%d:%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  ");
+               seq_printf(f, "LEASE  ");
                if (fl->fl_type & F_INPROGRESS)
-                       out += sprintf(out, "BREAKING  ");
+                       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 ",
+               seq_printf(f, "%s ",
                               (fl->fl_type & F_INPROGRESS)
                               ? (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, (long)f->private, "");
+
+       list_for_each_entry(bfl, &fl->fl_block, fl_block)
+               lock_get_status(f, bfl, (long)f->private, " ->");
+
+       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)
+{
+       lock_kernel();
+       f->private = (void *)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;
+       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);
-
-               if(pos >= offset+length)
-                       goto done;
-
-               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);
-
-                       if(pos >= offset+length)
-                               goto done;
-               }
-       }
-done:
+static void locks_stop(struct seq_file *f, void *v)
+{
        unlock_kernel();
-       *start = buffer;
-       if(q-buffer < length)
-               return (q-buffer);
-       return length;
 }
 
+static const struct seq_operations locks_seq_operations = {
+       .start  = locks_start,
+       .next   = locks_next,
+       .stop   = locks_stop,
+       .show   = locks_show,
+};
+
+static int locks_open(struct inode *inode, struct file *filp)
+{
+       return seq_open(filp, &locks_seq_operations);
+}
+
+static const struct file_operations proc_locks_operations = {
+       .open           = locks_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
+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
  *     @inode: the inode that is being read
@@ -2276,7 +2296,7 @@ static int __init filelock_init(void)
 {
        filelock_cache = kmem_cache_create("file_lock_cache",
                        sizeof(struct file_lock), 0, SLAB_PANIC,
-                       init_once, NULL);
+                       init_once);
        return 0;
 }