Re: [PATCH v5 bpf-next 2/3] mm/bpf: Add bpf_get_kmem_cache() kfunc

[Alexei Starovoitov]

On Tue, Oct 15, 2024 at 11:20 AM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
>
> On Mon, Oct 14, 2024 at 06:50:49PM -0700, Alexei Starovoitov wrote:
> > On Mon, Oct 14, 2024 at 11:13 AM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> > >
> > > Hi Alexei,
> > >
> > > On Fri, Oct 11, 2024 at 12:14:14PM -0700, Namhyung Kim wrote:
> > > > On Fri, Oct 11, 2024 at 11:35:27AM -0700, Alexei Starovoitov wrote:
> > > > > On Thu, Oct 10, 2024 at 4:25 PM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > The bpf_get_kmem_cache() is to get a slab cache information from a
> > > > > > virtual address like virt_to_cache().  If the address is a pointer
> > > > > > to a slab object, it'd return a valid kmem_cache pointer, otherwise
> > > > > > NULL is returned.
> > > > > >
> > > > > > It doesn't grab a reference count of the kmem_cache so the caller is
> > > > > > responsible to manage the access.  The returned point is marked as
> > > > > > PTR_UNTRUSTED.  And the kfunc has KF_RCU_PROTECTED as the slab object
> > > > > > might be protected by RCU.
> > > > >
> > > > > ...
> > > > > > +BTF_ID_FLAGS(func, bpf_get_kmem_cache, KF_RCU_PROTECTED)
> > > > >
> > > > > This flag is unnecessary. PTR_UNTRUSTED can point to absolutely any memory.
> > > > > In this case it likely points to a valid kmem_cache, but
> > > > > the verifier will guard all accesses with probe_read anyway.
> > > > >
> > > > > I can remove this flag while applying.
> > > >
> > > > Ok, I'd be happy if you would remove it.
> > >
> > > You will need to update the bpf_rcu_read_lock/unlock() in the test code
> > > (patch 3).  I can send v6 with that and Vlastimil's Ack if you want.
> >
> > Fixed all that while applying.
> >
> > Could you please follow up with an open-coded iterator version
> > of the same slab iterator ?
> > So that progs can iterate slabs as a normal for/while loop ?
>
> I'm not sure I'm following.  Do you want a new test program to iterate
> kmem_caches by reading list pointers manually?  How can I grab the
> slab_mutex then?

No.
See bpf_iter_task_new/_next/_destroy kfuncs and
commit c68a78ffe2cb ("bpf: Introduce task open coded iterator kfuncs").