Re: [PATCH 1/2] ima: Pre-parse the list of keyrings in a KEY_CHECK rule

[Nayna <nayna@linux.vnet.ibm.com>]

On 7/27/20 10:08 AM, Tyler Hicks wrote:
> The ima_keyrings buffer was used as a work buffer for strsep()-based
> parsing of the "keyrings=" option of an IMA policy rule. This parsing
> was re-performed each time an asymmetric key was added to a kernel
> keyring for each loaded policy rule that contained a "keyrings=" option.
>
> An example rule specifying this option is:
>
>   measure func=KEY_CHECK keyrings=a|b|c
>
> The rule says to measure asymmetric keys added to any of the kernel
> keyrings named "a", "b", or "c". The size of the buffer size was
> equal to the size of the largest "keyrings=" value seen in a previously
> loaded rule (5 + 1 for the NUL-terminator in the previous example) and
> the buffer was pre-allocated at the time of policy load.
>
> The pre-allocated buffer approach suffered from a couple bugs:
>
> 1) There was no locking around the use of the buffer so concurrent key
>     add operations, to two different keyrings, would result in the
>     strsep() loop of ima_match_keyring() to modify the buffer at the same
>     time. This resulted in unexpected results from ima_match_keyring()
>     and, therefore, could cause unintended keys to be measured or keys to
>     not be measured when IMA policy intended for them to be measured.
>
> 2) If the kstrdup() that initialized entry->keyrings in ima_parse_rule()
>     failed, the ima_keyrings buffer was freed and set to NULL even when a
>     valid KEY_CHECK rule was previously loaded. The next KEY_CHECK event
>     would trigger a call to strcpy() with a NULL destination pointer and
>     crash the kernel.
>
> Remove the need for a pre-allocated global buffer by parsing the list of
> keyrings in a KEY_CHECK rule at the time of policy load. The
> ima_rule_entry will contain an array of string pointers which point to
> the name of each keyring specified in the rule. No string processing
> needs to happen at the time of asymmetric key add so iterating through
> the list and doing a string comparison is all that's required at the
> time of policy check.
>
> In the process of changing how the "keyrings=" policy option is handled,
> a couple additional bugs were fixed:
>
> 1) The rule parser accepted rules containing invalid "keyrings=" values
>     such as "a|b||c", "a|b|", or simply "|".
>
> 2) The /sys/kernel/security/ima/policy file did not display the entire
>     "keyrings=" value if the list of keyrings was longer than what could
>     fit in the fixed size tbuf buffer in ima_policy_show().
>
> Fixes: 5c7bac9fb2c5 ("IMA: pre-allocate buffer to hold keyrings string")
> Fixes: 2b60c0ecedf8 ("IMA: Read keyrings= option from the IMA policy")
> Signed-off-by: Tyler Hicks <tyhicks@linux.microsoft.com>
> ---
>   security/integrity/ima/ima_policy.c | 138 +++++++++++++++++++---------
>   1 file changed, 93 insertions(+), 45 deletions(-)
>
> diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c
> index 07f033634b27..c328cfa0fc49 100644
> --- a/security/integrity/ima/ima_policy.c
> +++ b/security/integrity/ima/ima_policy.c
> @@ -59,6 +59,11 @@ enum policy_types { ORIGINAL_TCB = 1, DEFAULT_TCB };
>   
>   enum policy_rule_list { IMA_DEFAULT_POLICY = 1, IMA_CUSTOM_POLICY };
>   
> +struct ima_rule_opt_list {
> +	size_t count;
> +	char *items[];
> +};
> +
>   struct ima_rule_entry {
>   	struct list_head list;
>   	int action;
> @@ -78,7 +83,7 @@ struct ima_rule_entry {
>   		int type;	/* audit type */
>   	} lsm[MAX_LSM_RULES];
>   	char *fsname;
> -	char *keyrings; /* Measure keys added to these keyrings */
> +	struct ima_rule_opt_list *keyrings; /* Measure keys added to these keyrings */
>   	struct ima_template_desc *template;
>   };
>   
> @@ -206,10 +211,6 @@ static LIST_HEAD(ima_policy_rules);
>   static LIST_HEAD(ima_temp_rules);
>   static struct list_head *ima_rules = &ima_default_rules;
>   
> -/* Pre-allocated buffer used for matching keyrings. */
> -static char *ima_keyrings;
> -static size_t ima_keyrings_len;
> -
>   static int ima_policy __initdata;
>   
>   static int __init default_measure_policy_setup(char *str)
> @@ -253,6 +254,72 @@ static int __init default_appraise_policy_setup(char *str)
>   }
>   __setup("ima_appraise_tcb", default_appraise_policy_setup);
>   
> +static struct ima_rule_opt_list *ima_alloc_rule_opt_list(const substring_t *src)
> +{
> +	struct ima_rule_opt_list *opt_list;
> +	size_t count = 0;
> +	char *src_copy;
> +	char *cur, *next;
> +	size_t i;
> +
> +	src_copy = match_strdup(src);
> +	if (!src_copy)
> +		return NULL;

The caller of this function checks for IS_ERR(..) and not 
IS_ERR_OR_NULL(..). Shouldn't it return ERR_PTR(-EINVAL) instead of NULL ?

Thanks & Regards,

     - Nayna