Re: [PATCH v2] asymmetric_keys: check asymmetric_key_ids() for NULL before dereference

On Sat, May 02, 2026 at 09:33:29AM -0700, Weiming Shi wrote:
>
> diff --git a/crypto/asymmetric_keys/asymmetric_type.c b/crypto/asymmetric_keys/asymmetric_type.c
> index 16a7ae16593c..22f04656d529 100644
> --- a/crypto/asymmetric_keys/asymmetric_type.c
> +++ b/crypto/asymmetric_keys/asymmetric_type.c
> @@ -109,6 +109,8 @@ struct key *find_asymmetric_key(struct key *keyring,
>  	if (id_0 && id_1) {
>  		const struct asymmetric_key_ids *kids = asymmetric_key_ids(key);
>  
> +		if (!kids)
> +			goto reject;

This check is actually unnecessary because we've already matched
the key against the kid so it must be present.

I'd get rid of this check or perhaps add a comment instead.

>  		if (!kids->id[1]) {
>  			pr_debug("First ID matches, but second is missing\n");
>  			goto reject;
> diff --git a/crypto/asymmetric_keys/restrict.c b/crypto/asymmetric_keys/restrict.c
> index 86292965f493..ccf1084f720e 100644
> --- a/crypto/asymmetric_keys/restrict.c
> +++ b/crypto/asymmetric_keys/restrict.c
> @@ -243,10 +243,14 @@ static int key_or_keyring_common(struct key *dest_keyring,
>  			if (IS_ERR(key))
>  				key = NULL;
>  		} else if (trusted->type == &key_type_asymmetric) {
> +			const struct asymmetric_key_ids *kids;
>  			const struct asymmetric_key_id **signer_ids;
>  
> -			signer_ids = (const struct asymmetric_key_id **)
> -				asymmetric_key_ids(trusted)->id;
> +			kids = asymmetric_key_ids(trusted);
> +			if (!kids)
> +				goto skip_trusted;

Yes this is definitely buggy.

I think it was introduced by these two commits:

commit 3c58b2362ba828ee2970c66c6a6fd7b04fde4413
Author: David Howells <[email protected]>
Date:   Tue Oct 9 17:47:46 2018 +0100

    KEYS: Implement PKCS#8 RSA Private Key parser [ver #2]

and

commit 7e3c4d22083f6e7316c5229b6197ca2d5335aa35
Author: Mat Martineau <[email protected]>
Date:   Mon Jun 27 16:45:16 2016 -0700

    KEYS: Restrict asymmetric key linkage using a specific keychain

So the Fixes header should point to them.

> @@ -290,6 +294,7 @@ static int key_or_keyring_common(struct key *dest_keyring,
>  		}
>  	}
>  
> +skip_trusted:
>  	if (check_dest && !key) {
>  		/* See if the destination has a key that signed this one. */
>  		key = find_asymmetric_key(dest_keyring, sig->auth_ids[0],

I'm not sure continuing here is a good idea.  Having a private key
here makes no sense whatsoever and we should just bail out right
away.

I would recommend returning an error of some sort if kids is NULL.

David/Lukas/Ignat, any opinions?

Thanks,
-- 
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt