Re: [PATCH net-next] net: filter: rename 'struct sk_filter' to 'struct bpf_prog'

On Fri, Jul 25, 2014 at 6:00 AM, Daniel Borkmann [off-list ref] wrote:

On 07/25/2014 01:54 PM, Pablo Neira Ayuso wrote:

quoted

On Fri, Jul 25, 2014 at 01:25:35PM +0200, Daniel Borkmann wrote:

quoted

[ also Cc'ing Willem, Pablo ]

On 07/25/2014 10:04 AM, Alexei Starovoitov wrote:

quoted

'sk_filter' name is used as 'struct sk_filter', function sk_filter() and
as variable 'sk_filter', which makes code hard to read.
Also it's easily confused with 'struct sock_filter'
Rename 'struct sk_filter' to 'struct bpf_prog' to clarify semantics and
align the name with generic BPF use model.

Agreed, as we went for kernel/bpf/, renaming makes absolutely sense.

My nft socket filtering changes are accomodated into struct sk_filter,
and will still be, so I still need some generic name there...

All the parts from filter.c which is BPF's core engine have been moved
into kernel/bpf/ to get it ready for tracing et al, since there is not
always a socket context anymore. The *whole* infrastructure around struct
sk_filter is [e]BPF and used in non-net related contexts as well, whereas
nft socket filtering is *only* for sockets. Due to the socket-only specific
use case why doesn't it make more sense to have a union in struct sock
around sk_filter (or however we name it) and only allow one of the two
being loaded on a socket?

yep.
Adding nft specific things to struct sk_filter/bpf_prog is not correct,
since this struct is already part of seccomp and will be used
in net-less configurations. SK_RUN_FILTER() macro will also be
renamed into something like RUN_BPF_RPOG(). It's one and only
way to invoke eBPF programs. Adding nft selector cannot work,
since eBPF is used with generic context whereas nft is skb specific.
If you want to add nft filtering capabilities to sockets, you'd need
to add union around 'struct bpf_prog' inside 'struct sock', which will be
much cleaner way.