Re: [RFC] ima: check ima-policy's path

[Mimi Zohar <zohar@linux.ibm.com>]

On Tue, 2021-06-29 at 14:38 +0800, Tianxing Zhang wrote:
> Hi, I was reading the function ima_write_policy in ima/ima_fs.c when
> I find the issue:
> 
> > static ssize_t ima_write_policy(struct file *file, const char
> __user *buf,
> >                                              size_t datalen, loff_t
> *ppos)
> > {
> >         ...
> >
> >         if (data[0] == '/') {
> >                 result = ima_read_policy(data);
> >         } else if (ima_appraise & IMA_APPRAISE_POLICY) {
> >                 pr_err("signed policy file (specified as an
> absolute pathname) required\n");
> >                 integrity_audit_msg(AUDIT_INTEGRITY_STATUS, NULL,
> NULL,
> >                                              "policy_update",
> "signed policy required",
> >                                              1, 0);
> >         ...
> >         return result;
> > }
> 
> For the absolute path written by the user, we only check the prefix
> "/". Actually, we can echo an illegal path to the
> /sys/kernel/security/ima/policy, e.g. "/\rtest: ddddddddddddddddddd"
> to inject some logs into dmesg.
> 
> Then ima_read_policy is called to return error:
> 
> > static ssize_t ima_read_policy(char *path)
> > {
> >         ...
> >         rc = kernel_read_file_from_path(path, 0, &data, INT_MAX,
> NULL,
> >                                                      
> READING_POLICY);
> >         if (rc < 0) {
> >                 pr_err("Unable to open file: %s (%d)", path, rc);
> >                 return rc;
> >         }
> >         ...
> > }
> 
> In ima_read_policy, the illegal path would be logged into dmesg like
> this:
> 
> > ...
> > test: ddddddddddddddddddd (-2)/
> > test: ddddddddddddddddddd (-2)/
> > test: ddddddddddddddddddd (-2)/
> > test: ddddddddddddddddddd (-2)/
> 
> I suggest that we should check the path in ima_write_policy to make
> sure it's a valid one, at least literally.

Sure.  In the case that the path isn't valid, perhaps instead of
removing the message entirely, limit the number of messages emitted
using pr_err_once().

thanks,

Mimi