Re: [PATCH 2/2] configfs: unhash the dentry before dropping the item in rmdir

On 7/31/26 11:29 AM, Breno Leitao wrote:
> On Thu, Jul 30, 2026 at 11:30:25AM +0200, Vasileios Almpanis wrote:
>> configfs_get_config_item() treats a hashed dentry as proof that
>> sd->s_element is a live config_item.  configfs_rmdir() breaks that:
>> simple_rmdir() leaves the dentry hashed, the last reference to the item is
>> dropped right after, and the dentry is only unhashed by d_delete() once
>> ->rmdir() has returned.  configfs_symlink() resolves its target holding no
>> lock on it, so get_target() can land in that window:
>>
>>    BUG: KASAN: slab-use-after-free in config_item_get+0x26/0x90
>>     get_target fs/configfs/symlink.c:128 [inline]
>>     configfs_symlink+0x4ab/0x1030 fs/configfs/symlink.c:185
>>
>> Unhash in configfs_remove_dir(), while the item is still guaranteed to be
>> there.  A reference obtained just before that stays harmless, as
>> create_link() rechecks CONFIGFS_USET_DROPPING, already set by
>> configfs_detach_prep().  Both configfs_unregister_subsystem() paths
>> d_drop() after detaching, so this only makes rmdir match them.
> Is the claim that "a reference obtained just before that stays harmless,
> as create_link() rechecks CONFIGFS_USET_DROPPING" accurate?
I believe so.  Everything configfs_remove_dir() will now d_drop(), had
DROPPING set on its dirent beforehand by configfs_detach_prep(), and
once detach_prep() has succeeded the flag is never cleared again on
that subtree.

The race is closed by configfs_dirent_lock. create_link() checks DROPPING
and incs ->s_links in one locked section, and detach_prep() set DROPPING
and checks ->s_links in another. Either symlink gets ->s_links++ in 
first and
rmdir fails with -EBUSY before unhashing, or DROPPING is set first and
create_link() fails with -ENOENT.