Add ENOBUFS handling for unsolicited messages

This can happen when large burst of messages come all of a sudden, which
happen very easily when routing protocols are involved (e.g. BGP). The
current implementation incorrectly assumes that any failure to read from
the socket is akin to the socket closed. This is not the case.

This commit adds handling for this specific error by generating a
`NetlinkPayload::Overrun(_)` message that users receive on their
unsolicited message channel. Since this is just an additional message,
there is no breaking change for existing users and they are free to
ignore it if they do not want to handle it, or handle it by e.g. resyncing.

src/framed.rs

@@ -2,11 +2,7 @@
 
 use bytes::BytesMut;
 use std::{
-    fmt::Debug,
-    io,
-    marker::PhantomData,
-    pin::Pin,
-    task::{Context, Poll},
+    fmt::Debug, io, marker::PhantomData, mem::size_of, pin::Pin, task::{Context, Poll}
 };
 
 use futures::{Sink, Stream};
@@ -17,9 +13,13 @@ use crate::{
     sys::{AsyncSocket, SocketAddr},
 };
 use netlink_packet_core::{
-    NetlinkDeserializable, NetlinkMessage, NetlinkSerializable,
+    NetlinkDeserializable, NetlinkMessage, NetlinkSerializable, NetlinkHeader,
+    NLMSG_OVERRUN, NetlinkPayload,
 };
 
+/// Buffer overrun condition
+const ENOBUFS: i32 = 105;
+
 pub struct NetlinkFramed<T, S, C> {
     socket: S,
     // see https://doc.rust-lang.org/nomicon/phantom-data.html
@@ -68,6 +68,26 @@ where
 
             *in_addr = match ready!(socket.poll_recv_from(cx, reader)) {
                 Ok(addr) => addr,
+                // When receiving messages in multicast mode (i.e. we subscribed to
+                // notifications), the kernel will not wait for us to read datagrams before
+                // sending more. The receive buffer has a finite size, so once it is full (no
+                // more message can fit in), new messages will be dropped and recv calls will
+                // return `ENOBUFS`.
+                // This needs to be handled for applications to resynchronize with the contents
+                // of the kernel if necessary.
+                // We don't need to do anything special:
+                // - contents of the reader is still valid because we won't have partial messages
+                //   in there anyways (large enough buffer)
+                // - contents of the socket's internal buffer is still valid because the kernel
+                //   won't put partial data in it
+                Err(e) if e.raw_os_error() == Some(ENOBUFS) => {
+                    warn!("netlink socket buffer full");
+                    let mut hdr = NetlinkHeader::default();
+                    hdr.length = size_of::<NetlinkHeader>() as u32;
+                    hdr.message_type = NLMSG_OVERRUN;
+                    let msg = NetlinkMessage::new(hdr, NetlinkPayload::Overrun(Vec::new()));
+                    return Poll::Ready(Some((msg, SocketAddr::new(0, 0))));
+                }
                 Err(e) => {
                     error!("failed to read from netlink socket: {:?}", e);
                     return Poll::Ready(None);