Hello. On Thu, Nov 21, 2024 at 06:52:50PM GMT, Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote: > - kernfs_rename_ns() is only using kernfs_rename_lock if the parents are > different. All users users use either RCU or kernfs_rwsem. > - kernfs_fop_readdir() drops kernfs_root::kernfs_rwsem while holding a > reference to name and invoking dir_emit(). This has been changed and > lock is held. > - kernfs_notify_workfn() access kernfs_node::name without any > protection. Added kernfs_root::kernfs_rwsem for the iteration. > - kernfs_get_parent_dentry() acquires now kernfs_root::kernfs_rwsem > while accessing the parent node. > - kernfs_node_dentry() acquires now kernfs_root::kernfs_rwsem while > parent is accessed and the name looked up. Why is the kernfs_root::kernfs_rwsem newly R-taken? Shouldn't be RCU read section sufficient for those users? (Perhaps it's related to second observation I have -- why there is sometimes kernfs_rcu_get_parent() whereas there are other call sites with mere rcu_dereference(kn->parent)?)
Attachment:
signature.asc
Description: PGP signature
