Re: [PATCH bpf-next v3 01/16] bpf, netns: Handle multiple link attachments

[Jakub Sitnicki <jakub@cloudflare.com>]

On Thu, Jul 09, 2020 at 05:44 AM CEST, Andrii Nakryiko wrote:
> On Thu, Jul 2, 2020 at 2:24 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:
>>
>> Extend the BPF netns link callbacks to rebuild (grow/shrink) or update the
>> prog_array at given position when link gets attached/updated/released.
>>
>> This let's us lift the limit of having just one link attached for the new
>> attach type introduced by subsequent patch.
>>
>> No functional changes intended.
>>
>> Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
>> ---
>>
>> Notes:
>>     v3:
>>     - New in v3 to support multi-prog attachments. (Alexei)
>>
>>  include/linux/bpf.h        |  4 ++
>>  kernel/bpf/core.c          | 22 ++++++++++
>>  kernel/bpf/net_namespace.c | 88 +++++++++++++++++++++++++++++++++++---
>>  3 files changed, 107 insertions(+), 7 deletions(-)
>>
>> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
>> index 3d2ade703a35..26bc70533db0 100644
>> --- a/include/linux/bpf.h
>> +++ b/include/linux/bpf.h
>> @@ -928,6 +928,10 @@ int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs,
>>
>>  void bpf_prog_array_delete_safe(struct bpf_prog_array *progs,
>>                                 struct bpf_prog *old_prog);
>> +void bpf_prog_array_delete_safe_at(struct bpf_prog_array *array,
>> +                                  unsigned int index);
>> +void bpf_prog_array_update_at(struct bpf_prog_array *array, unsigned int index,
>> +                             struct bpf_prog *prog);
>>  int bpf_prog_array_copy_info(struct bpf_prog_array *array,
>>                              u32 *prog_ids, u32 request_cnt,
>>                              u32 *prog_cnt);
>> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
>> index 9df4cc9a2907..d4b3b9ee6bf1 100644
>> --- a/kernel/bpf/core.c
>> +++ b/kernel/bpf/core.c
>> @@ -1958,6 +1958,28 @@ void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
>>                 }
>>  }
>>
>> +void bpf_prog_array_delete_safe_at(struct bpf_prog_array *array,
>> +                                  unsigned int index)
>> +{
>> +       bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
>> +}
>> +
>> +void bpf_prog_array_update_at(struct bpf_prog_array *array, unsigned int index,
>> +                             struct bpf_prog *prog)
>
> it's a good idea to mention it in a comment for both delete_safe_at
> and update_at that slots with dummy entries are ignored.

I agree. These two need doc comments. update_at doesn't event hint that
this is not a regular update operation. Will add in v4.

>
> Also, given that index can be out of bounds, should these functions
> actually return error if the slot is not found?

That won't hurt. I mean, from bpf-netns PoV getting such an error would
indicate that there is a bug in the code that manages prog_array. But
perhaps other future users of this new prog_array API can benefit.

>
>> +{
>> +       struct bpf_prog_array_item *item;
>> +
>> +       for (item = array->items; item->prog; item++) {
>> +               if (item->prog == &dummy_bpf_prog.prog)
>> +                       continue;
>> +               if (!index) {
>> +                       WRITE_ONCE(item->prog, prog);
>> +                       break;
>> +               }
>> +               index--;
>> +       }
>> +}
>> +
>>  int bpf_prog_array_copy(struct bpf_prog_array *old_array,
>>                         struct bpf_prog *exclude_prog,
>>                         struct bpf_prog *include_prog,
>> diff --git a/kernel/bpf/net_namespace.c b/kernel/bpf/net_namespace.c
>> index 247543380fa6..6011122c35b6 100644
>> --- a/kernel/bpf/net_namespace.c
>> +++ b/kernel/bpf/net_namespace.c
>> @@ -36,11 +36,51 @@ static void netns_bpf_run_array_detach(struct net *net,
>>         bpf_prog_array_free(run_array);
>>  }
>>
>> +static unsigned int link_index(struct net *net,
>> +                              enum netns_bpf_attach_type type,
>> +                              struct bpf_netns_link *link)
>> +{
>> +       struct bpf_netns_link *pos;
>> +       unsigned int i = 0;
>> +
>> +       list_for_each_entry(pos, &net->bpf.links[type], node) {
>> +               if (pos == link)
>> +                       return i;
>> +               i++;
>> +       }
>> +       return UINT_MAX;
>
> Why not return a negative error, if the slot is not found? Feels a bit
> unusual as far as error reporting goes.

Returning uint played well with the consumer of link_index() return
value, that is bpf_prog_array_update_at(). update at takes an index into
the array, which must not be negative.

But I don't have strong feelings toward it. Will switch to -ENOENT in
v4.

>
>> +}
>> +
>
> [...]