Re: [PATCH bpf v2] af_unix: Disable MSG_OOB handling for sockets in sockmap/sockhash

[Jakub Sitnicki <jakub@cloudflare.com>]

On Wed, Jun 26, 2024 at 12:19 PM +02, Michal Luczaj wrote:
> On 6/24/24 16:15, Jakub Sitnicki wrote:
>> On Sun, Jun 23, 2024 at 12:25 AM +02, Michal Luczaj wrote:
>>> AF_UNIX socket tracks the most recent OOB packet (in its receive queue)
>>> with an `oob_skb` pointer. BPF redirecting does not account for that: when
>>> an OOB packet is moved between sockets, `oob_skb` is left outdated. This
>>> results in a single skb that may be accessed from two different sockets.
>>>
>>> Take the easy way out: silently drop MSG_OOB data targeting any socket that
>>> is in a sockmap or a sockhash. Note that such silent drop is akin to the
>>> fate of redirected skb's scm_fp_list (SCM_RIGHTS, SCM_CREDENTIALS).
>>>
>>> For symmetry, forbid MSG_OOB in unix_bpf_recvmsg().
>>>
>>> Suggested-by: Kuniyuki Iwashima <kuniyu@amazon.com>
>>> Fixes: 314001f0bf92 ("af_unix: Add OOB support")
>>> Signed-off-by: Michal Luczaj <mhal@rbox.co>
>>> ---
>> 
>> [+CC Cong who authored ->read_skb]
>> 
>> I'm guessing you have a test program that you're developing the fix
>> against. Would you like to extend the test case for sockmap redirect
>> from unix stream [1] to incorporate it?
>> 
>> Sadly unix_inet_redir_to_connected needs a fix first because it
>> hardcodes sotype to SOCK_DGRAM.
>
> Ugh, my last two replies got silently dropped by vger. Is there any way to
> tell what went wrong?

Not sure if it was vger or lore archive. Your reply hit my inbox but is
nowhere to be found in the archive:

https://lore.kernel.org/r/4bac0a8a-eeaa-48ef-aeba-2a6e73c0b982@rbox.co

I think we can reach out to Konstantin Ryabitsev at
konstantin@linuxfoundation.org. AFAIK he maintains the lore.kernel.org
archive.

> So, again, sure, I'll extend the sockmap redirect test.

Appreciate the help with adding a regression test, if time allows.
Fixes are of course very welcome even without them.

> And regarding Rao's comment, I took a look and I think sockmap'ed TCP OOB
> does indeed act the same way. I'll try to add that into selftest as well.n

Right, it does sound like we're not clearing the offset kept in
tcp_sock::urg_data when skb is redirected.