Re: [PATCH 4/6] kernfs: don't take kernfs_iattr_rwsem to publish KERNFS_REMOVING
From: T.J. Mercier
Date: Fri Nov 20 2026 - 13:58:07 EST
On Wed, Nov 18, 2026 at 2:02 PM Shakeel Butt <shakeel.butt@xxxxxxxxx> wrote:
>
> __kernfs_remove() takes kernfs_iattr_rwsem for writing while it marks
> a subtree KERNFS_REMOVING, inside kernfs_rwsem. Commit 507d8ce13f5b
> ("kernfs: Don't set_nlink for directories being removed") added it so
> that a refresh would not set the i_nlink of a directory whose removal
> has started. The refresh no longer sets i_nlink, so the lock is not
> needed.
>
> The flag is only set here, under kernfs_rwsem. Inode setup reads it
> under kernfs_rwsem. getattr reads it without and reports the old link
> count or 0, either of which is right for a stat() racing the removal.
>
> Drop the lock, mark the store and the load, and read the flags once in
> kernfs_dir_nlink(), as the type bits share the word.
>
> kernfs_iattr_rwsem write acquisitions (lock_stat), before -> after:
>
> 5 dummy netdevs created and removed 1110 -> 740
> 100 cgroup mkdirs and 100 rmdirs 4500 -> 3000
>
> Signed-off-by: Shakeel Butt <shakeel.butt@xxxxxxxxx>
> ---
> fs/kernfs/dir.c | 10 ++++++----
> fs/kernfs/inode.c | 7 +++++--
> 2 files changed, 11 insertions(+), 6 deletions(-)
>
> diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c
> index ef23eac684c6..890aca5d8d70 100644
> --- a/fs/kernfs/dir.c
> +++ b/fs/kernfs/dir.c
> @@ -1604,15 +1604,17 @@ static void __kernfs_remove(struct kernfs_node *kn)
>
> pr_debug("kernfs %s: removing\n", kernfs_rcu_name(kn));
>
> - /* prevent new usage by marking all nodes removing and deactivating */
> - down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
> + /*
> + * Prevent new usage by marking all nodes removing and deactivating.
> + * KERNFS_REMOVING is only set here, under kernfs_rwsem.
> + * kernfs_dir_nlink() reads it without that lock.
> + */
> pos = NULL;
> while ((pos = kernfs_next_descendant_post(pos, kn))) {
> - pos->flags |= KERNFS_REMOVING;
> + WRITE_ONCE(pos->flags, pos->flags | KERNFS_REMOVING);
> if (kernfs_active(pos))
> atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
> }
> - up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
>
> /* deactivate and unlink the subtree node-by-node */
> do {
> diff --git a/fs/kernfs/inode.c b/fs/kernfs/inode.c
> index 2f463c0354a5..ca234f69882b 100644
> --- a/fs/kernfs/inode.c
> +++ b/fs/kernfs/inode.c
> @@ -165,9 +165,12 @@ static inline void set_inode_attr(struct inode *inode,
> */
> static bool kernfs_dir_nlink(struct kernfs_node *kn, unsigned int *nlink)
> {
> - if (kernfs_type(kn) != KERNFS_DIR)
> + /* KERNFS_REMOVING can be set concurrently; read the flags once. */
> + unsigned short flags = READ_ONCE(kn->flags);
Hi Shakeel,
I'd like to ask for some code comments / documentation.
This lockless read relies on two rules for kn->flags... the type bits
and flags set before publication never change, and the other flags are
only changed with kernfs_rwsem held for write (except KERNFS_NS).
Parts of this are explained in a comment in kernfs_dop_revalidate()
but not on the field, and that comment doesn't cover KERNFS_HIDDEN or
KERNFS_SUICIDAL/SUICIDED.
Since this series adds a lockless reader that depends on these rules,
how about a comment on kernfs_node->flags like:
/*
* Type bits, flags set before publication, and KERNFS_NS are stable;
the rest change under
* kernfs_rwsem (write). Lockless readers of changing flags must use
READ_ONCE(), taking
* one snapshot if they also test stable bits.
*/
> +
> + if ((flags & KERNFS_TYPE_MASK) != KERNFS_DIR)
> return false;
> - *nlink = (kn->flags & KERNFS_REMOVING) ? 0 :
> + *nlink = (flags & KERNFS_REMOVING) ? 0 :
> READ_ONCE(kn->dir.subdirs) + 2;
> return true;
> }
> --
> 2.53.1-Meta
>