virtio-net: disable delayed refill when pausing rx
authorBui Quang Minh <minhquangbui99@gmail.com>
Thu, 17 Apr 2025 07:28:03 +0000 (14:28 +0700)
committerJakub Kicinski <kuba@kernel.org>
Wed, 23 Apr 2025 01:29:13 +0000 (18:29 -0700)
commit4bc12818b363bd30f0f7348dd9ab077290a637ae
treef8aa194c91aeeb6455495d5908fda4977784965e
parentb7f0ee992adf601aa00c252418266177eb7ac2bc
virtio-net: disable delayed refill when pausing rx

When pausing rx (e.g. set up xdp, xsk pool, rx resize), we call
napi_disable() on the receive queue's napi. In delayed refill_work, it
also calls napi_disable() on the receive queue's napi.  When
napi_disable() is called on an already disabled napi, it will sleep in
napi_disable_locked while still holding the netdev_lock. As a result,
later napi_enable gets stuck too as it cannot acquire the netdev_lock.
This leads to refill_work and the pause-then-resume tx are stuck
altogether.

This scenario can be reproducible by binding a XDP socket to virtio-net
interface without setting up the fill ring. As a result, try_fill_recv
will fail until the fill ring is set up and refill_work is scheduled.

This commit adds virtnet_rx_(pause/resume)_all helpers and fixes up the
virtnet_rx_resume to disable future and cancel all inflights delayed
refill_work before calling napi_disable() to pause the rx.

Fixes: 413f0271f396 ("net: protect NAPI enablement with netdev_lock()")
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Bui Quang Minh <minhquangbui99@gmail.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20250417072806.18660-2-minhquangbui99@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/virtio_net.c