Hi Mimi,
On 7/26/21 9:33 AM, Mimi Zohar wrote:
Hi Tushar,
On Sat, 2021-07-24 at 00:25 -0700, Tushar Sugandhi wrote:
Hi Mimi,
Missing from the document is a way of validating the template data.
For example, in the original case of file measurements, the template
data contains the file hash, which can be recalculated or verified
against an allow list.
Other than re-calculating the template data digest based on the
template data, and verifying it against the template data digest in the
measurement list, would an attestation server be able to verify the
template data itself?
Yes.
In the context of device-mapper, EVENT_DATA for 'table_load' would
contain the key-value pairs for various targets in the table
(crypt, verity, integrity etc.) which the attestation servers
should be able to verify against the allowed/expected
key-value pairs specified in the attestation policy.
To avoid bloating the IMA log with same data from table_load again,
we only measure hash of the loaded table in the EVENT_DATA -
when there is a state change for DM device.
e.g. when EVENT_NAME is 'device_resume', 'table_clear',
'device_remove' etc.
Since the table clear-text is already present in the EVENT_DATA
buffer for 'table_load', and is available to attestation servers,
verifying the corresponding hash values in the
EVENT_DATA in the subsequent DM events should be possible for
the attestation servers.
Please let us know if you need further info.
For regular files with signatures, the file signature is verified
against the file hash, both contained within the template data. For
the SELinux "critical-data",
commit 2554a48f4437 ("selinux: measure state and policy capabilities")
contains that information. Missing from this patch set is information
on how the attestation server could verify the DM critical data.
Does the DM record contain everything needed for the attestion server
to verify the template record? Are things like the hash algorithm hard
coded?
Thanks for the feedback.
I think you have a point (on documenting hash algo and version in the
IMA record).
For the individual targets (crypt/linear etc.) the target version
is already recorded during table_load [1].
So for new attribute add/update/delete to those targets in future,
they should be accompanied by bumping the relevant target version.
So that the attestation servers would know what attributes to expect,
based on the target's version number.
So individual targets are already taken care of.
But for overall DM records I think some versioning would be useful
to the attestation servers, to determine what to expect for the
attributes that are at device level (i.e. not part of any individual
targets e.g. "minor_count", "num_targets" etc.).
Further, for the DM records that are not "table_load" - what hashing
algo is used to hash active/inactive table is the implementation detail
specific to DM, as Mike called out; and it is orthogonal to IMA's hash
algo.
But I think it would be useful to record the active/inactive table's
hash algo as part of the overall DM record template, so that attestation
servers would know to how to interpret the relevant data.
(i.e. "active_table_hash", "inactive_table_hash" in [2]).
Both the changes are simple enough.
I will prototype and propose them as incremental patches along with
other changes (pointed to dm-5.15 branch as per Mike's suggestion.)
Thanks for your suggestions again.
Regards,
Tushar
[1]
+10 a8c5ff755561c7a28146389d1514c318592af49a ima-buf
sha256:4d73481ecce5eadba8ab084640d85bb9ca899af4d0a122989252a76efadc5b72
+table_load
+name=linear1,uuid=,major=253,minor=0,minor_count=1,num_targets=1;
+target_index=0,target_begin=0,target_len=2,target_name=linear,target_version=1.4.0,device_name=7:0,start=512;
+table_load
+name=test-crypt,uuid=,major=253,minor=0,minor_count=1,num_targets=1;
+target_index=0,target_begin=0,target_len=1953125,target_name=crypt,target_version=1.23.0,
[2]
+device_remove
+device_active_metadata=name=linear1,uuid=,major=253,minor=0,minor_count=1,num_targets=4;
+device_inactive_metadata=name=linear1,uuid=,major=253,minor=0,minor_count=1,num_targets=2;
+active_table_hash=4d73481ecce5eadba8ab084640d85bb9ca899af4d0a122989252a76efadc5b72,
+inactive_table_hash=5596cc857b0e887fd0c5d58dc6382513284596b07f09fd37efae2da224bd521d,remove_all=n;
+current_device_capacity=8;
thanks,
Mimi
