Re: [PATCH 1/1] security: Propagate universal pointer data in bpf hooks

[Blaise Boscaccy <bboscaccy@linux.microsoft.com>]

Paul Moore <paul@paul-moore.com> writes:

> On Wed, Feb 26, 2025 at 2:06 AM Song Liu <song@kernel.org> wrote:
>> On Tue, Feb 25, 2025 at 4:31 PM Blaise Boscaccy
>> <bboscaccy@linux.microsoft.com> wrote:
>> >
>> > Certain bpf syscall subcommands are available for usage from both
>> > userspace and the kernel. LSM modules or eBPF gatekeeper programs may
>> > need to take a different course of action depending on whether or not
>> > a BPF syscall originated from the kernel or userspace.
>> >
>> > Additionally, some of the bpf_attr struct fields contain pointers to
>> > arbitrary memory. Currently the functionality to determine whether or
>> > not a pointer refers to kernel memory or userspace memory is exposed
>> > to the bpf verifier, but that information is missing from various LSM
>> > hooks.
>> >
>> > Here we augment the LSM hooks to provide this data, by simply passing
>> > the corresponding universal pointer in any hook that contains already
>> > contains a bpf_attr struct that corresponds to a subcommand that may
>> > be called from the kernel.
>>
>> I think this information is useful for LSM hooks.
>
> I've only looked at it quickly, but so far it seems reasonable.  I'm
> going to take a closer look today.
>
>> Question: Do we need a full bpfptr_t for these hooks, or just a boolean
>> "is_kernel or not"?
>
> I may be misunderstanding the patch, but what if we swapped the
> existing 'union bpf_attr' parameter for a 'bpfptr_t' parameter?  That
> would allow for both kernel and usermode pointers, complete with a
> 'is_kernel' flag; or am I missing something (likely)?
>
> -- 
> paul-moore.com

bpfptr_t is just a typedef for a sockptr_t, which contains a void
pointer and bool, so if we replaced bpf_attr with it, we might lose a
bit of type safety going that route.

In syscall.c a most of the subcommand handlers have a

static int bpf_foo(union bpf_attr *attr, bpfptr_t uattr);

pattern that is used. I was trying to mimic for this patch.

The actual parts where the is_kernel flag gets used currently, is for
pointer chasing/copy stuff, e.g.

make_bpfptr(attr->insns, uattr.is_kernel)
make_bpfptr(attr->license, uattr.is_kernel)
make_bpfptr(attr->fd_array, uattr.is_kernel)

and subcommand structs may contain multiple pointers.

-blaise