vhost/vsock: Allocate nonlinear SKBs for handling large receive buffers
authorWill Deacon <will@kernel.org>
Thu, 17 Jul 2025 09:01:14 +0000 (10:01 +0100)
committerMichael S. Tsirkin <mst@redhat.com>
Fri, 1 Aug 2025 13:11:09 +0000 (09:11 -0400)
commitab9aa2f3afc2713c14f6c4c6b90c9a0933b837f1
tree3ae368ca78f336b957f769afa2e023e8eea09366
parentfac6b82e0f3eaca33c8c67ec401681b21143ae17
vhost/vsock: Allocate nonlinear SKBs for handling large receive buffers

When receiving a packet from a guest, vhost_vsock_handle_tx_kick()
calls vhost_vsock_alloc_linear_skb() to allocate and fill an SKB with
the receive data. Unfortunately, these are always linear allocations and
can therefore result in significant pressure on kmalloc() considering
that the maximum packet size (VIRTIO_VSOCK_MAX_PKT_BUF_SIZE +
VIRTIO_VSOCK_SKB_HEADROOM) is a little over 64KiB, resulting in a 128KiB
allocation for each packet.

Rework the vsock SKB allocation so that, for sizes with page order
greater than PAGE_ALLOC_COSTLY_ORDER, a nonlinear SKB is allocated
instead with the packet header in the SKB and the receive data in the
fragments. Finally, add a debug warning if virtio_vsock_skb_rx_put() is
ever called on an SKB with a non-zero length, as this would be
destructive for the nonlinear case.

Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Will Deacon <will@kernel.org>
Message-Id: <20250717090116.11987-8-will@kernel.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
drivers/vhost/vsock.c
include/linux/virtio_vsock.h