[PATCH] IMA: set a default value for unknown digsig algorithms

[PATCH] IMA: set a default value for unknown digsig algorithms

[THOBY Simon]

When adding new protections against writing invalid data in
the security.ima xattr, I erroneously expected ima_get_hash_algo()
to always return a valid 'enum hash_algo', but it turns out it trusts
the user-supplied digital signatures and return it without any checks.
It didn't affect process_measurement() because that function
(indirectly) calls into ima_alloc_atfm() that fallback silently
on the default hash algorithm, but it did affect ima_inode_setxattr
as that new function didn't perform a bounds check.

Update ima_get_hash_algo() to always return a valid hash algorithm,
defaulting on 'ima_hash_algo' when the user-supplied value inside
the xattr is invalid.

This patch was successfully tested by syszbot, see
https://syzkaller.appspot.com/bug?extid=e8bafe7b82c739eaf153.

Signed-off-by: THOBY Simon <Simon.THOBY@viveris.fr>
Reported-by: syzbot+e8bafe7b82c739eaf153@syzkaller.appspotmail.com
---
 security/integrity/ima/ima_appraise.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/security/integrity/ima/ima_appraise.c b/security/integrity/ima/ima_appraise.c
index 8f1eb7ef041e..dbba51583e7c 100644
--- a/security/integrity/ima/ima_appraise.c
+++ b/security/integrity/ima/ima_appraise.c
@@ -185,7 +185,8 @@ enum hash_algo ima_get_hash_algo(const struct evm_ima_xattr_data *xattr_value,
 	switch (xattr_value->type) {
 	case EVM_IMA_XATTR_DIGSIG:
 		sig = (typeof(sig))xattr_value;
-		if (sig->version != 2 || xattr_len <= sizeof(*sig))
+		if (sig->version != 2 || xattr_len <= sizeof(*sig)
+		    || sig->hash_algo >= HASH_ALGO__LAST)
 			return ima_hash_algo;
 		return sig->hash_algo;
 		break;
-- 
2.31.1

Re: [PATCH] IMA: set a default value for unknown digsig algorithms

[Lakshmi Ramasubramanian]

On 8/20/2021 5:18 AM, THOBY Simon wrote:
> When adding new protections against writing invalid data in
> the security.ima xattr, I erroneously expected ima_get_hash_algo()
> to always return a valid 'enum hash_algo', but it turns out it trusts
> the user-supplied digital signatures and return it without any checks.
> It didn't affect process_measurement() because that function
> (indirectly) calls into ima_alloc_atfm() that fallback silently
> on the default hash algorithm, but it did affect ima_inode_setxattr
> as that new function didn't perform a bounds check.
> 
> Update ima_get_hash_algo() to always return a valid hash algorithm,
> defaulting on 'ima_hash_algo' when the user-supplied value inside
> the xattr is invalid.
> 
> This patch was successfully tested by syszbot, see
> https://syzkaller.appspot.com/bug?extid=e8bafe7b82c739eaf153.
> 
> Signed-off-by: THOBY Simon <Simon.THOBY@viveris.fr>
> Reported-by: syzbot+e8bafe7b82c739eaf153@syzkaller.appspotmail.com
> ---
>   security/integrity/ima/ima_appraise.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)

Reviewed-by: Lakshmi Ramasubramanian <nramas@linux.microsoft.com>

  -lakshmi

> 
> diff --git a/security/integrity/ima/ima_appraise.c b/security/integrity/ima/ima_appraise.c
> index 8f1eb7ef041e..dbba51583e7c 100644
> --- a/security/integrity/ima/ima_appraise.c
> +++ b/security/integrity/ima/ima_appraise.c
> @@ -185,7 +185,8 @@ enum hash_algo ima_get_hash_algo(const struct evm_ima_xattr_data *xattr_value,
>   	switch (xattr_value->type) {
>   	case EVM_IMA_XATTR_DIGSIG:
>   		sig = (typeof(sig))xattr_value;
> -		if (sig->version != 2 || xattr_len <= sizeof(*sig))
> +		if (sig->version != 2 || xattr_len <= sizeof(*sig)
> +		    || sig->hash_algo >= HASH_ALGO__LAST)
>   			return ima_hash_algo;
>   		return sig->hash_algo;
>   		break;
> 

Re: [PATCH] IMA: set a default value for unknown digsig algorithms

[Mimi Zohar]

Hi Simon,

On Fri, 2021-08-20 at 12:18 +0000, THOBY Simon wrote:
> When adding new protections against writing invalid data in
> the security.ima xattr, I erroneously expected ima_get_hash_algo()
> to always return a valid 'enum hash_algo', but it turns out it trusts
> the user-supplied digital signatures and return it without any checks.
> It didn't affect process_measurement() because that function
> (indirectly) calls into ima_alloc_atfm() that fallback silently
> on the default hash algorithm, but it did affect ima_inode_setxattr
> as that new function didn't perform a bounds check.

Please update patch description something like:

The new function validate_hash_algo() assumed that ima_get_hash_algo()
always return a valid 'enum hash_algo', but returned the user-supplied hash algorithm from the digital signature without any checks.

> 
> Update ima_get_hash_algo() to always return a valid hash algorithm,
> defaulting on 'ima_hash_algo' when the user-supplied value inside
> the xattr is invalid.
> 
> This patch was successfully tested by syszbot, see
> https://syzkaller.appspot.com/bug?extid=e8bafe7b82c739eaf153.

Thank you for the pointer.
> 
> Signed-off-by: THOBY Simon <Simon.THOBY@viveris.fr>
> Reported-by: syzbot+e8bafe7b82c739eaf153@syzkaller.appspotmail.com
Fixes: 50f742dd9147 ("IMA: block writes of the security.ima xattr with
unsupported algorithms")

thanks,

Mimi

> ---
>  security/integrity/ima/ima_appraise.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/security/integrity/ima/ima_appraise.c b/security/integrity/ima/ima_appraise.c
> index 8f1eb7ef041e..dbba51583e7c 100644
> --- a/security/integrity/ima/ima_appraise.c
> +++ b/security/integrity/ima/ima_appraise.c
> @@ -185,7 +185,8 @@ enum hash_algo ima_get_hash_algo(const struct evm_ima_xattr_data *xattr_value,
>  	switch (xattr_value->type) {
>  	case EVM_IMA_XATTR_DIGSIG:
>  		sig = (typeof(sig))xattr_value;
> -		if (sig->version != 2 || xattr_len <= sizeof(*sig))
> +		if (sig->version != 2 || xattr_len <= sizeof(*sig)
> +		    || sig->hash_algo >= HASH_ALGO__LAST)
>  			return ima_hash_algo;
>  		return sig->hash_algo;
>  		break;