[PATCH v24 04/25] IMA: avoid label collisions with stacked LSMs

[Casey Schaufler]

On 2/14/2021 10:21 AM, Mimi Zohar wrote:
> Hi Casey,
>
> On Tue, 2021-01-26 at 08:40 -0800, Casey Schaufler wrote:
>> Integrity measurement may filter on security module information
>> and needs to be clear in the case of multiple active security
>> modules which applies. Provide a boot option ima_rules_lsm= to
>> allow the user to specify an active securty module to apply
>> filters to. If not specified, use the first registered module
>> that supports the audit_rule_match() LSM hook. Allow the user
>> to specify in the IMA policy an lsm= option to specify the
>> security module to use for a particular rule.
> Thanks, Casey.
>
> (This patch description line length seems short.)
>
>> Signed-off-by: Casey Schaufler <casey at schaufler-ca.com>
>> To: Mimi Zohar <zohar at linux.ibm.com>
>> To: linux-integrity at vger.kernel.org
>> ---
>>  Documentation/ABI/testing/ima_policy |  8 +++-
>>  security/integrity/ima/ima_policy.c  | 64 ++++++++++++++++++++++------
>>  2 files changed, 57 insertions(+), 15 deletions(-)
>>
>> diff --git a/Documentation/ABI/testing/ima_policy b/Documentation/ABI/testing/ima_policy
>> index e35263f97fc1..a7943d40466f 100644
>> --- a/Documentation/ABI/testing/ima_policy
>> +++ b/Documentation/ABI/testing/ima_policy
>> @@ -25,7 +25,7 @@ Description:
>>  			base:	[[func=] [mask=] [fsmagic=] [fsuuid=] [uid=]
>>  				[euid=] [fowner=] [fsname=]]
>>  			lsm:	[[subj_user=] [subj_role=] [subj_type=]
>> -				 [obj_user=] [obj_role=] [obj_type=]]
>> +				 [obj_user=] [obj_role=] [obj_type=] [lsm=]]
> "[lsm=]" either requires all LSM rules types (e.g. {subj/obj}_user,
> role, type) to be exactly the same for multiple LSMs or all of the LSM
> rule types are applicable to only a single LSM.  Supporting multiple
> LSMs with exactly the same LSM labels doesn't seem worth the effort.  
> Keep it simple - a single rule, containing any LSM rule types, is
> applicable to a single LSM.

Thank you. I will add this.

>
>>  			option:	[[appraise_type=]] [template=] [permit_directio]
>>  				[appraise_flag=] [keyrings=]
>>  		  base:
>> @@ -114,6 +114,12 @@ Description:
>>
>>  			measure subj_user=_ func=FILE_CHECK mask=MAY_READ
>>
>> +		It is possible to explicitly specify which security
>> +		module a rule applies to using lsm=.  If the security
>> +		modules specified is not active on the system the rule
>> +		will be rejected.  If lsm= is not specified the first
>> +		security module registered on the system will be assumed.
>> +
>>  		Example of measure rules using alternate PCRs::
>>
>>  			measure func=KEXEC_KERNEL_CHECK pcr=4
>> diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c
>> index 8002683003e6..de72b719c90c 100644
>> --- a/security/integrity/ima/ima_policy.c
>> +++ b/security/integrity/ima/ima_policy.c
>> @@ -82,6 +82,7 @@ struct ima_rule_entry {
>>  		void *rules[LSMBLOB_ENTRIES]; /* LSM file metadata specific */
>>  		char *args_p;	/* audit value */
>>  		int type;	/* audit type */
>> +		int which_lsm; /* which of the rules to use */
>>  	} lsm[MAX_LSM_RULES];
> Even if we wanted to support multiple LSMs within the same rule having
> both "rules[LSMBLOB_ENTRIES]" and "which_lsm" shouldn't be necessary.  
> The LSMBLOB_ENTRIES should already identify the LSM.
>
> To support a single LSM per policy rule, "which_lsm" should be defined
> outside of lsm[MAX_LSM_RULES].  This will simplify the rest of the code
> (e.g. matching/freeing rules).
>
> 	int which_lsm;          /* which of the rules to use */
> 	struct {
>                 void *rule;        /* LSM file metadata specific */
>                 char *args_p;   /* audit value */
>                 int type;       /* audit type */
>         } lsm[MAX_LSM_RULES];

You're right, that is better. I'll incorporate the change.

>
>
>>  	char *fsname;
>>  	struct ima_rule_opt_list *keyrings; /* Measure keys added to these keyrings */
>> @@ -90,17 +91,15 @@ struct ima_rule_entry {
>>
>>  /**
>>   * ima_lsm_isset - Is a rule set for any of the active security modules
>> - * @rules: The set of IMA rules to check
>> + * @entry: the rule entry to examine
>> + * @lsm_rule: the specific rule type in question
>>   *
>> - * If a rule is set for any LSM return true, otherwise return false.
>> + * If a rule is set return true, otherwise return false.
>>   */
>> -static inline bool ima_lsm_isset(void *rules[])
>> +static inline bool ima_lsm_isset(struct ima_rule_entry *entry, int lsm_rule)
>>  {
>> -	int i;
>> -
>> -	for (i = 0; i < LSMBLOB_ENTRIES; i++)
>> -		if (rules[i])
>> -			return true;
>> +	if (entry->lsm[lsm_rule].rules[entry->lsm[lsm_rule].which_lsm])
>> +		return true;
> If each IMA policy rule is limited to a specific LSM, then the test
> would be "entry->which_lsm".

Which would be an improvement.

>
>>  	return false;
>>  }
>>
>> @@ -273,6 +272,20 @@ static int __init default_appraise_policy_setup(char *str)
>>  }
>>  __setup("ima_appraise_tcb", default_appraise_policy_setup);
>>
>> +static int ima_rule_lsm __ro_after_init;
>> +
>> +static int __init ima_rule_lsm_init(char *str)
>> +{
>> +	ima_rule_lsm = lsm_name_to_slot(str);
>> +	if (ima_rule_lsm < 0) {
>> +		ima_rule_lsm = 0;
>> +		pr_err("rule lsm \"%s\" not registered", str);
>> +	}
>> +
>> +	return 1;
>> +}
>> +__setup("ima_rule_lsm=", ima_rule_lsm_init);
> The patch description refers to "ima_rules_lsm=".  Please update one or
> the other.

ima_rules_lsm seem to be more accurate. I'll fix it.

>
> thanks,
>
> Mimi

Thanks for the review and recommendations.