On Fri, 2020-12-04 at 08:05 +0000, Roberto Sassu wrote:

quoted

From: Mimi Zohar [mailto:zohar@linux.ibm.com]
Sent: Thursday, December 3, 2020 9:43 PM
Hi Roberto,

On Wed, 2020-11-11 at 10:22 +0100, Roberto Sassu wrote:

quoted

When a file is being created, LSMs can set the initial label with the
inode_init_security hook. If no HMAC key is loaded, the new file will have
LSM xattrs but not the HMAC.

Unfortunately, EVM will deny any further metadata operation on new

files,

quoted

as evm_protect_xattr() will always return the INTEGRITY_NOLABEL error.

This

quoted

would limit the usability of EVM when only a public key is loaded, as
commands such as cp or tar with the option to preserve xattrs won't work.

Ignoring this error won't be an issue if no HMAC key is loaded, as the
inode is locked until the post hook, and EVM won't calculate the HMAC on
metadata that wasn't previously verified. Thus this patch checks if an
HMAC key is loaded and if not, ignores INTEGRITY_NOLABEL.

I'm not sure what problem this patch is trying to solve.
evm_protect_xattr() is only called by evm_inode_setxattr() and
evm_inode_removexattr(), which first checks whether
EVM_ALLOW_METADATA_WRITES is enabled.

The idea is to also support EVM verification when only a public key
is loaded. An advantage to do that is that for example we can prevent
accidental metadata changes when the signature is portable.

Right, there are a couple of  scenarios.  Let's be more specific as to
which scenario this patch is addressing.

- a public key is loaded and EVM_ALLOW_METADATA_WRITES is enabled,
- a public key is loaded and EVM_ALLOW_METADATA_WRITES is disabled,
- an HMAC key is loaded

For the first and last case, this patch shouldn't be necessary.  Only
the second case, with EVM_ALLOW_METADATA_WRITES disabled, probably does
not work.  I would claim that is working as designed.

thanks,

Mimi