When default groups are nested, lockdep raises a warning since it sees a
lock recursion of class I_MUTEX_CHILD in populate_groups().
However, this lock recursion is just a variant of the I_MUTEX_PARENT ->
I_MUTEX_CHILD dependency, which is ok in the VFS, and is already checked when
creating the first level of default groups.

This patch silences lockdep with nested default groups, by hiding the mutex
locks of populate_groups() from lockdep when the considered config_group is
itself a default group. The mutex is not hidden for a non-default group, so that
the lock dependency remains checked, in case things change in some future.

Signed-off-by: Louis Rilling <[EMAIL PROTECTED]>
---
 fs/configfs/dir.c |   17 +++++++++++++++++
 1 file changed, 17 insertions(+)


Index: b/fs/configfs/dir.c
===================================================================
--- a/fs/configfs/dir.c 2008-05-20 18:19:35.000000000 +0200
+++ b/fs/configfs/dir.c 2008-05-20 18:19:41.000000000 +0200
@@ -535,6 +535,8 @@ static int populate_groups(struct config
        int i;
 
        if (group->default_groups) {
+               struct configfs_dirent *sd = dentry->d_fsdata;
+               int turn_lockdep_off = (sd->s_type & CONFIGFS_USET_DEFAULT);
                /*
                 * FYI, we're faking mkdir here
                 * I'm not sure we need this semaphore, as we're called
@@ -544,7 +546,18 @@ static int populate_groups(struct config
                 * That said, taking our i_mutex is closer to mkdir
                 * emulation, and shouldn't hurt.
                 */
+               /* For a default group, we hide this mutex from lockdep since:
+                * 1/ This is a case of I_MUTEX_PARENT -> I_MUTEX_CHILD
+                *    dependency;
+                * 2/ This dependency was already checked when creating the
+                *    parent of this group;
+                * 3/ Lockdep does not handle such safe recursion.
+                */
+               if (turn_lockdep_off)
+                       lockdep_off();
                mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD);
+               if (turn_lockdep_off)
+                       lockdep_on();
 
                for (i = 0; group->default_groups[i]; i++) {
                        new_group = group->default_groups[i];
@@ -554,7 +567,11 @@ static int populate_groups(struct config
                                break;
                }
 
+               if (turn_lockdep_off)
+                       lockdep_off();
                mutex_unlock(&dentry->d_inode->i_mutex);
+               if (turn_lockdep_off)
+                       lockdep_on();
        }
 
        if (ret)

-- 
Dr Louis Rilling                        Kerlabs
Skype: louis.rilling                    Batiment Germanium
Phone: (+33|0) 6 80 89 08 23            80 avenue des Buttes de Coesmes
http://www.kerlabs.com/                 35700 Rennes


_______________________________________________
Ocfs2-devel mailing list
[email protected]
http://oss.oracle.com/mailman/listinfo/ocfs2-devel

Reply via email to