* Re: [PATCH] KEYS: protect construct_key type dereference with key semaphore
2026-11-21 11:28 [PATCH] KEYS: protect construct_key type dereference with key semaphore Tristan Madani
@ 2026-11-21 11:38 ` sashiko-bot
0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-21 11:38 UTC (permalink / raw)
To: Tristan Madani, David Howells, Jarkko Sakkinen
Cc: Paul Moore, James Morris, Serge E . Hallyn, keyrings,
linux-security-module, Tristan Madani
Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Medium] KEYS: ineffective key->sem locking around request_key read in
construct_key()
--
commit ac4e3e194cb3618952a6681a5209a4099b22f9f6
Author: Tristan Madani <tristan@talencesecurity.com>
KEYS: protect construct_key type dereference with key semaphore
This commit takes key->sem for reading around the key->type->request_key
lookup in construct_key().
> diff --git a/security/keys/request_key.c b/security/keys/request_key.c
> index fa2bb9f2f5381..7e9fe4796319c 100644
> --- a/security/keys/request_key.c
> +++ b/security/keys/request_key.c
> @@ -241,8 +241,10 @@ static int construct_key(struct key *key, const void *callout_info,
>
> /* Make the call */
> actor = call_sbin_request_key;
> + down_read(&key->sem);
> if (key->type->request_key)
> actor = key->type->request_key;
> + up_read(&key->sem);
>
> ret = actor(authkey, aux);
>
[Severity: Medium]
Does taking key->sem only while reading key->type->request_key actually
protect the actor(authkey, aux) call from module unload, or can a
concurrent type removal happen here at all?
Unlike 925c84e3cc21 ("KEYS: Protect the read method check against type
removal"), 51bcd09b9ffc ("KEYS: Protect the update method check against
type removal") and bab48a104447 ("KEYS: Protect the pkey method checks
against type removal"), which hold key->sem across the method execution,
construct_key() drops key->sem before calling actor(authkey, aux). If
unregister_key_type() could run concurrently, key_garbage_collector()
could take down_write(&key->sem) right after up_read(&key->sem), retype
the key to key_type_dead, and allow the module text to be freed before
actor(authkey, aux) runs.
More importantly, construct_key() is only reached via
request_key_and_link(struct key_type *type, ...), which already requires
the key type to stay registered across the entire call and accesses
type->match_preparse before construct_key() and type->match_free after
it without key->sem:
security/keys/request_key.c:request_key_and_link() {
...
key = construct_key_and_link(&ctx, callout_info, callout_len,
aux, dest_keyring, flags);
}
error_free:
if (type->match_free)
type->match_free(&ctx.match_data);
...
}
On the sys_request_key() syscall path, key_type_lookup() holds
key_types_sem for reading across request_key_and_link() until
key_type_put(), blocking unregister_key_type() before key_gc_keytype()
can start, while in-kernel callers pass struct key_type * directly and
must keep the type registered for the duration of the request.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261121112853.2175119-1-tristmd@gmail.com?part=1
^ permalink raw reply [flat|nested] 2+ messages in thread