Re: [PATCH] unix: properly account for FDs passed over unix sockets

[Hannes Frederic Sowa <hannes@stressinduktion.org>]

On 29.12.2015 21:35, Willy Tarreau wrote:
> On Tue, Dec 29, 2015 at 03:48:45PM +0100, Hannes Frederic Sowa wrote:
>> On 28.12.2015 15:14, Willy Tarreau wrote:
>>> It is possible for a process to allocate and accumulate far more FDs than
>>> the process' limit by sending them over a unix socket then closing them
>>> to keep the process' fd count low.
>>>
>>> This change addresses this problem by keeping track of the number of FDs
>>> in flight per user and preventing non-privileged processes from having
>>> more FDs in flight than their configured FD limit.
>>>
>>> Reported-by: socketpair@gmail.com
>>> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
>>> Signed-off-by: Willy Tarreau <w@1wt.eu>
>>
>> Thanks for the patch!
>>
>> I think this does not close the DoS attack completely as we duplicate
>> fds if the reader uses MSG_PEEK on the unix domain socket and thus
>> clones the fd. Have I overlooked something?
>
> I didn't know this behaviour. However, then the fd remains in flight, right ?
> So as long as it's not removed from the queue, the sender cannot add more
> than its FD limit. I may be missing something obvious though :-/

Yes, it remains in flight.

The MSG_PEEK code should not be harmful and the patch is good as is. I 
first understood from the published private thread, that it is possible 
for a program to exceed the rlimit of fds. But the DoS is only by 
keeping the fds in flight and not attaching them to any program.

__alloc_fd, called on the receiver side, does check for the rlimit 
maximum anyway, so I don't see a loophole anymore:

Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>

Another idea would be to add the amount of memory used to manage the fds 
to sock_rmem/wmem but I don't see any advantages or disadvantages.

Thanks!