Re: [PATCH net-next v7 08/16] quic: add path management

On Tue, Jan 20, 2026 at 10:34 AM Xin Long <[email protected]> wrote:
>
> On Tue, Jan 20, 2026 at 9:13 AM Paolo Abeni <[email protected]> wrote:
> >
> > On 1/15/26 4:11 PM, Xin Long wrote:
> > > @@ -0,0 +1,524 @@
> > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > +/* QUIC kernel implementation
> > > + * (C) Copyright Red Hat Corp. 2023
> > > + *
> > > + * This file is part of the QUIC kernel implementation
> > > + *
> > > + * Initialization/cleanup for QUIC protocol support.
> > > + *
> > > + * Written or modified by:
> > > + *    Xin Long <[email protected]>
> > > + */
> > > +
> > > +#include <net/udp_tunnel.h>
> > > +#include <linux/quic.h>
> > > +
> > > +#include "common.h"
> > > +#include "family.h"
> > > +#include "path.h"
> > > +
> > > +static int (*quic_path_rcv)(struct sock *sk, struct sk_buff *skb, u8 err);
> >
> > It's unclear why an indirect call is needed here. At least some
> > explanation is needed in the commit message, possibly you could call
> > directly a static function.
> >
> This patchset makes the path subcomponent more independent from the core
> implementation. Aside from a few shared helper functions, it doesn't
> rely on code outside the subcomponent, in particular socket.c and
> packet.c.
>
> Other subcomponents, such as stream, connid, pnspace, cong,
> crypto, common, and family follow the same approach. They expose
> APIs to the core QUIC logic and don’t see the main process.
>
> Will leave an explanation here.
>
To achieve this, instead of introducing an indirect call, we can use
an 'extern' declaration of quic_packet_rcv() once it is introduced.
Before that point, the code simply falls back to calling kfree_skb(skb).

Will update it.

Thanks.

> > > +
> > > +static int quic_udp_rcv(struct sock *sk, struct sk_buff *skb)
> > > +{
> > > +     memset(skb->cb, 0, sizeof(skb->cb));
> > > +     QUIC_SKB_CB(skb)->seqno = -1;
> > > +     QUIC_SKB_CB(skb)->time = quic_ktime_get_us();
> > > +
> > > +     skb_pull(skb, sizeof(struct udphdr));
> > > +     skb_dst_force(skb);
> > > +     quic_path_rcv(sk, skb, 0);
> > > +     return 0;
> >
> > Why not:
> >         return quic_path_rcv(sk, skb, 0);
> > ?
> I checked the udp tunnel users:
>
> - bareudp: bareudp_udp_encap_recv()
> - vxlan: vxlan_rcv()
> - geneve: geneve_udp_encap_recv()
> - tipc: tipc_udp_recv()
> - sctp: sctp_udp_rcv()
>
> they all are returning 0 in .encap_udp(), I think because they all
> take care of the
> skb freeing in their err path already.
>
> is it safe to return error to UDP stack if the skb is already freed?
> Do you know?
>
> >
> > > +static struct quic_udp_sock *quic_udp_sock_create(struct sock *sk, union quic_addr *a)
> > > +{
> > > +     struct udp_tunnel_sock_cfg tuncfg = {};
> > > +     struct udp_port_cfg udp_conf = {};
> > > +     struct net *net = sock_net(sk);
> > > +     struct quic_uhash_head *head;
> > > +     struct quic_udp_sock *us;
> > > +     struct socket *sock;
> > > +
> > > +     us = kzalloc(sizeof(*us), GFP_KERNEL);
> > > +     if (!us)
> > > +             return NULL;
> > > +
> > > +     quic_udp_conf_init(sk, &udp_conf, a);
> > > +     if (udp_sock_create(net, &udp_conf, &sock)) {
> > > +             pr_debug("%s: failed to create udp sock\n", __func__);
> > > +             kfree(us);
> > > +             return NULL;
> > > +     }
> > > +
> > > +     tuncfg.encap_type = 1;
> > > +     tuncfg.encap_rcv = quic_udp_rcv;
> > > +     tuncfg.encap_err_lookup = quic_udp_err;
> > > +     setup_udp_tunnel_sock(net, sock, &tuncfg);
> > > +
> > > +     refcount_set(&us->refcnt, 1);
> > > +     us->sk = sock->sk;
> > > +     memcpy(&us->addr, a, sizeof(*a));
> > > +     us->bind_ifindex = sk->sk_bound_dev_if;
> > > +
> > > +     head = quic_udp_sock_head(net, ntohs(a->v4.sin_port));
> > > +     hlist_add_head(&us->node, &head->head);
> > > +     INIT_WORK(&us->work, quic_udp_sock_put_work);
> > > +
> > > +     return us;
> > > +}
> > > +
> > > +static bool quic_udp_sock_get(struct quic_udp_sock *us)
> > > +{
> > > +     return refcount_inc_not_zero(&us->refcnt);
> > > +}
> > > +
> > > +static void quic_udp_sock_put(struct quic_udp_sock *us)
> > > +{
> > > +     if (refcount_dec_and_test(&us->refcnt))
> > > +             queue_work(quic_wq, &us->work);
> >
> > Why using a workqueue here? AFAICS all the caller are in process
> > context. Is that to break a possible deadlock due to nested mutex?
> > Likely a comment on the refcount/locking scheme would help.
> >
> quic_udp_sock_put() will also be called in the RX path via
> quic_path_unbind() for the connection migration (after changing
> to a new path.), which is in the patchset-2.
>
> I will leave a comment for an explanation here.
>
> Thanks.