On Wed, 9 Jun 2021 at 10:52, Ian Kent <raven@xxxxxxxxxx> wrote: > > Now the kernfs_rwsem read lock is held for kernfs_refresh_inode() and > the i_lock taken to protect inode updates there can be some contention > introduced when .permission() is called with concurrent path walks in > progress. > > Since .permission() is called frequently during path walks it's worth > checking if the update is actually needed before taking the lock and > performing the update. > > Signed-off-by: Ian Kent <raven@xxxxxxxxxx> Reviewed-by: Miklos Szeredi <mszeredi@xxxxxxxxxx>
