Re: [PATCH ghak8 ALT4 V4 0/3] audit: show more information for entries with anonymous parents

[Paul Moore]

On Fri, Feb 16, 2018 at 3:23 AM, Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> On 2018-02-15 17:15, Paul Moore wrote:
>> On Mon, Feb 12, 2018 at 12:02 AM, Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
>> > More than one filesystem was causing hundreds to thousands of null PATH
>> > records to be associated with the *init_module SYSCALL records on a few
>> > modules with corresponding audit syscall rules.
>> >
>> > This patchset adds extra information to those PATH records to provide
>> > insight into what is generating them, including a partial pathname,
>> > fstype field, and two new filetypes that indicate the pathname isn't
>> > anchored at the root of the task's root filesystem.
>> >
>> > Richard Guy Briggs (3):
>> >   audit: show partial pathname for entries with anonymous parents
>> >   audit: append new fstype field for anonymous PATH records
>> >   audit: add new filetypes CREATE_ANON and PARENT_ANON
>>
>> The more I look at this, the more I prefer your original approach that
>> prefixed the relative pathname with the fstype.  Yes, I do realize
>> that you sort of work around that by including the fstype as a new
>> field in the PATH records, but we're still stuck with those odd
>> relative/un-rooted name fields.
>
> They are signalled as being unrooted by the ANON filetypes.  And now
> that you mention it, should fail the ausearch-test since it isn't a "full
> path", as claimed is necessary in ghak70 (so I don't see why the
> KERN_MODULE name= record/field fails that test).

Yes.  I still prefer your original approach.

>> Further, I don't recall ever hearing a good reason why the original
>> approach wasn't acceptable to Steve's userspace.  I know he did make
>> some very last minute hand-wavy comments, but none of those made any
>> sense to me; I don't understand why Steve's audit record parser is
>> even looking in the pathname string.
>>
>> I'm going to park these patches in limbo for the time being.
>
> Can you give me an idea how long that might be?

If you need an answer right now, consider it to be "indefinitely".

-- 
paul moore
www.paul-moore.com