Re: [PATCH net] mptcp: fix uninitialized local_id in syncookie MP_JOIN reconstruction

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The `request_bkup` bitfield in `mptcp_subflow_request_sock` is left uninitialized during syncookie MP_JOIN reconstruction, causing stale data to dictate the subflow backup state.
--

--- Patch [1]: [PATCH net] mptcp: fix uninitialized local_id in syncookie MP_JOIN reconstruction ---
commit ac471f0880c8ac59556dc5bd33e8a83b1849726c
Author: Harshit Varu <[email protected]>

mptcp: fix uninitialized local_id in syncookie MP_JOIN reconstruction

This commit restores the local_id field from the saved cookie entry when
rebuilding the request socket for an MP_JOIN 4th-ACK handled under SYN
cookies. This prevents the subflow from using stale slab memory as its
address-ID, protecting the path manager's bookkeeping.

> diff --git a/net/mptcp/syncookies.c b/net/mptcp/syncookies.c
> index 7f22526346a7e..b5cac57011221 100644
> --- a/net/mptcp/syncookies.c
> +++ b/net/mptcp/syncookies.c
> @@ -118,6 +118,7 @@ bool mptcp_token_join_cookie_init_state(struct mptcp_subflow_request_sock *subfl
>  	subflow_req->local_nonce = e->local_nonce;
>  	subflow_req->backup = e->backup;
>  	subflow_req->remote_id = e->join_id;
> +	subflow_req->local_id = e->local_id;

[Severity: High]
This is a pre-existing issue, but does this code also leave the request_bkup
bitfield uninitialized during syncookie reconstruction?

While this commit fixes local_id, the request_bkup bitfield in
mptcp_subflow_request_sock doesn't appear to be saved in
mptcp_join_store_state() or restored here.

Since the request_sock slab is SLAB_TYPESAFE_BY_RCU and memory isn't zeroed
on allocation, subflow_init_req() zeroes some adjacent bitfields but misses
request_bkup:

net/mptcp/subflow.c:subflow_init_req() {
    ...
    subflow_req->mp_capable = 0;
    subflow_req->mp_join = 0;
    subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
    ...
}

When the final subflow context is cloned, it copies this uninitialized state
directly from the request socket:

net/mptcp/subflow.c:subflow_ulp_clone() {
    ...
    new_ctx->backup = subflow_req->backup;
    new_ctx->request_bkup = subflow_req->request_bkup;
    ...
}

Could this stale memory cause the MPTCP stack to erroneously treat primary
paths as backup paths (or vice versa)?

>  	subflow_req->token = e->token;
>  	subflow_req->msk = msk;
>  	spin_unlock_bh(&join_entry_locks[i]);

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1