Re: [PATCH] cifs: verify signature only for valid responses

[Enzo Matsumiya <ematsumiya@suse.de>]

Hi Tom,

On 09/17, Tom Talpey wrote:
<snip>
>Wait, we process the message *before* we check the signature??? Apart
>from inspecting the MID and verifying it's a response to a request we
>made, there isn't a lot to cause such an error. See 3.2.5.1.3.

You're right. By processing I actually meant "parsing" done right after
receive, but even that doesn't have many failure spots.

I found that the mids with STATUS_END_OF_FILE are being discarded,
apparently as per 3.2.5.11:

   If the Status field of the SMB2 header of the response indicates an
   error, the client MUST return the received status code to the calling
   application.

What I found is that mid->callback() (smb2_readv_callback()) was being
called from another thread, so even though the mid had been dequeued by
mid->receive() earlier, smb2_readv_callback() was treating it as a
valid (non-NULL), existing (mid_state == MID_RESPONSE_RECEIVED) mid.

 From this perspective, it makes sense to me to skip the signature
verification when the mid wasn't supposed to be there in the first
place, but if we consider that other messages with status !=
STATUS_SUCCESS have their signatures correctly computed (apparently),
then I'd guess there's something wrong with computing signatures for
STATUS_END_OF_FILE responses.

Sent this just now:
https://lore.kernel.org/linux-cifs/20220918235442.2981-1-ematsumiya@suse.de/T/#u

I'd appreciate your, and Cc'd folks', feedback.


Cheers,

Enzo