Re: [syzbot] [mptcp?] general protection fault in proc_scheduler

From: Matthieu Baerts
Date: Sun Jan 05 2025 - 12:03:49 EST


Hi Eric,

On 05/01/2025 17:52, Eric Dumazet wrote:
> On Sun, Jan 5, 2025 at 12:29 PM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>>
>> On Sun, Jan 05, 2025 at 09:32:36AM +0100, Eric Dumazet wrote:
>>
>>> According to grep, we have many other places directly reading
>>> current->nsproxy->net_ns
>>> For instance in net/sctp/sysctl.c
>>> Should we change them all ?
>>
>> Depends - do you want their contents match the netns of opener (as,
>> AFAICS, for ipv4 sysctls) or that of the reader?
>
> I am only worried that a malicious user could crash the host with
> current kernels,
> not about this MPTP crash, but all unaware users of current->nsproxy
> in sysctl handlers.
>
> Back to MPTCP :
>
> Using the convention used in other mptcp sysctls like (enabled,
> add_addr_timeout,
> checksum_enabled, allow_join_initial_addr_port...) is better for consistency.

Indeed, I can do the modifications to stop using current->nsproxy in
MPTCP. I can do the same in SCTP.

Do you plan to send your patch modifying proc_sysctl.c? It is just to
know if I should mark my patches as fixes, and split them to ease the
backports -- each helper using current->nsproxy has been introduced in
different commits -- or if I can send them to net-next instead.

Cheers,
Matt
--
Sponsored by the NGI0 Core fund.