diff mbox series

fsverity: skip PKCS#7 parser when keyring is empty

Message ID 20230801050714.28974-1-ebiggers@kernel.org (mailing list archive)
State Superseded
Headers show
Series fsverity: skip PKCS#7 parser when keyring is empty | expand

Commit Message

Eric Biggers Aug. 1, 2023, 5:07 a.m. UTC 
From: Eric Biggers <ebiggers@google.com>

If an fsverity builtin signature is given for a file but the
".fs-verity" keyring is empty, there's no real reason to run the PKCS#7
parser.  Skip this to avoid the PKCS#7 attack surface when builtin
signature support is configured into the kernel but is not being used.

This is a hardening improvement, not a fix per se, but I've added
Fixes and Cc stable to get it out to more users.

Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/verity/signature.c | 24 ++++++++++++++++++++----
 1 file changed, 20 insertions(+), 4 deletions(-)


base-commit: 456ae5fe9b448f44ebe98b391a3bae9c75df465e

Comments

Victor Hsieh Aug. 1, 2023, 4:07 p.m. UTC | #1 
Should the whole use of "d" be moved into the else block?


On Mon, Jul 31, 2023 at 10:09 PM Eric Biggers <ebiggers@kernel.org> wrote:
>
> From: Eric Biggers <ebiggers@google.com>
>
> If an fsverity builtin signature is given for a file but the
> ".fs-verity" keyring is empty, there's no real reason to run the PKCS#7
> parser.  Skip this to avoid the PKCS#7 attack surface when builtin
> signature support is configured into the kernel but is not being used.
>
> This is a hardening improvement, not a fix per se, but I've added
> Fixes and Cc stable to get it out to more users.
>
> Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> Cc: stable@vger.kernel.org
> Signed-off-by: Eric Biggers <ebiggers@google.com>
> ---
>  fs/verity/signature.c | 24 ++++++++++++++++++++----
>  1 file changed, 20 insertions(+), 4 deletions(-)
>
> diff --git a/fs/verity/signature.c b/fs/verity/signature.c
> index b95acae64eac6..f6668d92d8151 100644
> --- a/fs/verity/signature.c
> +++ b/fs/verity/signature.c
> @@ -70,10 +70,26 @@ int fsverity_verify_signature(const struct fsverity_info *vi,
>         d->digest_size = cpu_to_le16(hash_alg->digest_size);
>         memcpy(d->digest, vi->file_digest, hash_alg->digest_size);
>
> -       err = verify_pkcs7_signature(d, sizeof(*d) + hash_alg->digest_size,
> -                                    signature, sig_size, fsverity_keyring,
> -                                    VERIFYING_UNSPECIFIED_SIGNATURE,
> -                                    NULL, NULL);
> +       if (fsverity_keyring->keys.nr_leaves_on_tree == 0) {
> +               /*
> +                * The ".fs-verity" keyring is empty, due to builtin signatures
> +                * being supported by the kernel but not actually being used.
> +                * In this case, verify_pkcs7_signature() would always return an
> +                * error, usually ENOKEY.  It could also be EBADMSG if the
> +                * PKCS#7 is malformed, but that isn't very important to
> +                * distinguish.  So, just skip to ENOKEY to avoid the attack
> +                * surface of the PKCS#7 parser, which would otherwise be
> +                * reachable by any task able to execute FS_IOC_ENABLE_VERITY.
> +                */
> +               err = -ENOKEY;
> +       } else {
> +               err = verify_pkcs7_signature(d,
> +                                            sizeof(*d) + hash_alg->digest_size,
> +                                            signature, sig_size,
> +                                            fsverity_keyring,
> +                                            VERIFYING_UNSPECIFIED_SIGNATURE,
> +                                            NULL, NULL);
> +       }
>         kfree(d);
>
>         if (err) {
>
> base-commit: 456ae5fe9b448f44ebe98b391a3bae9c75df465e
> --
> 2.41.0
>
Jarkko Sakkinen Aug. 1, 2023, 6:53 p.m. UTC | #2 
On Tue Aug 1, 2023 at 8:07 AM EEST, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
>
> If an fsverity builtin signature is given for a file but the
> ".fs-verity" keyring is empty, there's no real reason to run the PKCS#7
> parser.  Skip this to avoid the PKCS#7 attack surface when builtin
> signature support is configured into the kernel but is not being used.
>
> This is a hardening improvement, not a fix per se, but I've added
> Fixes and Cc stable to get it out to more users.
>
> Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> Cc: stable@vger.kernel.org
> Signed-off-by: Eric Biggers <ebiggers@google.com>
> ---
>  fs/verity/signature.c | 24 ++++++++++++++++++++----
>  1 file changed, 20 insertions(+), 4 deletions(-)
>
> diff --git a/fs/verity/signature.c b/fs/verity/signature.c
> index b95acae64eac6..f6668d92d8151 100644
> --- a/fs/verity/signature.c
> +++ b/fs/verity/signature.c
> @@ -70,10 +70,26 @@ int fsverity_verify_signature(const struct fsverity_info *vi,
>  	d->digest_size = cpu_to_le16(hash_alg->digest_size);
>  	memcpy(d->digest, vi->file_digest, hash_alg->digest_size);
>  
> -	err = verify_pkcs7_signature(d, sizeof(*d) + hash_alg->digest_size,
> -				     signature, sig_size, fsverity_keyring,
> -				     VERIFYING_UNSPECIFIED_SIGNATURE,
> -				     NULL, NULL);
> +	if (fsverity_keyring->keys.nr_leaves_on_tree == 0) {
> +		/*
> +		 * The ".fs-verity" keyring is empty, due to builtin signatures
> +		 * being supported by the kernel but not actually being used.
> +		 * In this case, verify_pkcs7_signature() would always return an
> +		 * error, usually ENOKEY.  It could also be EBADMSG if the
> +		 * PKCS#7 is malformed, but that isn't very important to
> +		 * distinguish.  So, just skip to ENOKEY to avoid the attack
> +		 * surface of the PKCS#7 parser, which would otherwise be
> +		 * reachable by any task able to execute FS_IOC_ENABLE_VERITY.
> +		 */
> +		err = -ENOKEY;
> +	} else {
> +		err = verify_pkcs7_signature(d,
> +					     sizeof(*d) + hash_alg->digest_size,
> +					     signature, sig_size,
> +					     fsverity_keyring,
> +					     VERIFYING_UNSPECIFIED_SIGNATURE,
> +					     NULL, NULL);
> +	}
>  	kfree(d);
>  
>  	if (err) {
>
> base-commit: 456ae5fe9b448f44ebe98b391a3bae9c75df465e
> -- 
> 2.41.0

Acked-by: Jarkko Sakkinen <jarkko@kernel.org>

BR, Jarkko
Eric Biggers Aug. 2, 2023, 4:15 a.m. UTC | #3 
On Tue, Aug 01, 2023 at 09:07:44AM -0700, Victor Hsieh wrote:
> Should the whole use of "d" be moved into the else block?

In v2, changed to use an early return instead.

- Eric
diff mbox series

Patch

diff --git a/fs/verity/signature.c b/fs/verity/signature.c
index b95acae64eac6..f6668d92d8151 100644
--- a/fs/verity/signature.c
+++ b/fs/verity/signature.c
@@ -70,10 +70,26 @@  int fsverity_verify_signature(const struct fsverity_info *vi,
 	d->digest_size = cpu_to_le16(hash_alg->digest_size);
 	memcpy(d->digest, vi->file_digest, hash_alg->digest_size);
 
-	err = verify_pkcs7_signature(d, sizeof(*d) + hash_alg->digest_size,
-				     signature, sig_size, fsverity_keyring,
-				     VERIFYING_UNSPECIFIED_SIGNATURE,
-				     NULL, NULL);
+	if (fsverity_keyring->keys.nr_leaves_on_tree == 0) {
+		/*
+		 * The ".fs-verity" keyring is empty, due to builtin signatures
+		 * being supported by the kernel but not actually being used.
+		 * In this case, verify_pkcs7_signature() would always return an
+		 * error, usually ENOKEY.  It could also be EBADMSG if the
+		 * PKCS#7 is malformed, but that isn't very important to
+		 * distinguish.  So, just skip to ENOKEY to avoid the attack
+		 * surface of the PKCS#7 parser, which would otherwise be
+		 * reachable by any task able to execute FS_IOC_ENABLE_VERITY.
+		 */
+		err = -ENOKEY;
+	} else {
+		err = verify_pkcs7_signature(d,
+					     sizeof(*d) + hash_alg->digest_size,
+					     signature, sig_size,
+					     fsverity_keyring,
+					     VERIFYING_UNSPECIFIED_SIGNATURE,
+					     NULL, NULL);
+	}
 	kfree(d);
 
 	if (err) {