Re: [PATCH v2] ext4: fix use-after-free in update_super_work when racing with umount

[Jan Kara]

On Thu 19-03-26 15:36:50, Jiayuan Chen wrote:
> From: Jiayuan Chen <jiayuan.chen@xxxxxxxxxx>
> 
> Commit b98535d09179 ("ext4: fix bug_on in start_this_handle during umount
> filesystem") moved ext4_unregister_sysfs() before flushing s_sb_upd_work
> to prevent new error work from being queued via /proc/fs/ext4/xx/mb_groups
> reads during unmount. However, this introduced a use-after-free because
> update_super_work calls ext4_notify_error_sysfs() -> sysfs_notify() which
> accesses the kobject's kernfs_node after it has been freed by kobject_del()
> in ext4_unregister_sysfs():
> 
>   update_super_work                ext4_put_super
>   -----------------                --------------
>                                    ext4_unregister_sysfs(sb)
>                                      kobject_del(&sbi->s_kobj)
>                                        __kobject_del()
>                                          sysfs_remove_dir()
>                                            kobj->sd = NULL
>                                          sysfs_put(sd)
>                                            kernfs_put()  // RCU free
>   ext4_notify_error_sysfs(sbi)
>     sysfs_notify(&sbi->s_kobj)
>       kn = kobj->sd              // stale pointer
>       kernfs_get(kn)             // UAF on freed kernfs_node
>                                    ext4_journal_destroy()
>                                      flush_work(&sbi->s_sb_upd_work)
> 
> Instead of reordering the teardown sequence, fix this by making
> ext4_notify_error_sysfs() detect that sysfs has already been torn down
> by checking s_kobj.state_in_sysfs, and skipping the sysfs_notify() call
> in that case. A dedicated mutex (s_error_notify_mutex) serializes
> ext4_notify_error_sysfs() against kobject_del() in ext4_unregister_sysfs()
> to prevent TOCTOU races where the kobject could be deleted between the
> state_in_sysfs check and the sysfs_notify() call.
> 
> Fixes: b98535d09179 ("ext4: fix bug_on in start_this_handle during umount filesystem")
> Cc: Jiayuan Chen <jiayuan.chen@xxxxxxxxx>
> Suggested-by: Jan Kara <jack@xxxxxxx>
> Signed-off-by: Jiayuan Chen <jiayuan.chen@xxxxxxxxxx>

Thanks for the patch! Automated AI review
(https://sashiko.dev/#/patchset/20260319073651.79209-1-jiayuan.chen%40linux.dev)

has noticed two small issues with this:

> diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
> index d2ecc1026c0c..f2505988e010 100644
> --- a/fs/ext4/sysfs.c
> +++ b/fs/ext4/sysfs.c
> @@ -597,7 +597,10 @@ static const struct kobj_type ext4_feat_ktype = {
>  
>  void ext4_notify_error_sysfs(struct ext4_sb_info *sbi)
>  {
> -	sysfs_notify(&sbi->s_kobj, NULL, "errors_count");
> +	mutex_lock(&sbi->s_error_notify_mutex);
> +	if (sbi->s_kobj.state_in_sysfs)
> +		sysfs_notify(&sbi->s_kobj, NULL, "errors_count");
> +	mutex_unlock(&sbi->s_error_notify_mutex);
>  }
>  
>  static struct kobject *ext4_root;
> @@ -610,6 +613,7 @@ int ext4_register_sysfs(struct super_block *sb)
>  	int err;
>  
>  	init_completion(&sbi->s_kobj_unregister);
> +	mutex_init(&sbi->s_error_notify_mutex);
>  	err = kobject_init_and_add(&sbi->s_kobj, &ext4_sb_ktype, ext4_root,
>  				   "%s", sb->s_id);
>  	if (err) {

The initialization of s_error_notify_mutex should happen early in
ext4_fill_super() as ext4_notify_error_sysfs() can be called rather early
before ext4_register_sysfs() is called.

Also we should protect kobject_init_and_add() with s_error_notify_mutex to
handle the case where ext4_notify_error_sysfs() is racing with
ext4_register_sysfs().

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR