On Tue, 7 Oct 2003 15:12:06 +0400 (MSD)
kuznet@ms2.inr.ac.ru wrote:
> > Rather netlink_set_err(), which walks through socket table.
> > I think it is to be some new function, which sets overrun bit in socket
> > state as well.
>
> The second sentence is wrong. When we cannot allocate skb to send,
> it is enough to call netlink_set_err(), overrun bit is set only on
> shortage of receiver rcvbuf.
But I don't understand why we must walk socket table. If only
one socket is going to receive this netlink_ack() built packet,
why need we set error on any other socket?
For example, I think this patch would solve all the problems.
--- net/netlink/af_netlink.c.~1~ Tue Oct 7 04:04:55 2003
+++ net/netlink/af_netlink.c Tue Oct 7 04:12:07 2003
@@ -898,8 +898,17 @@
size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
skb = alloc_skb(size, GFP_KERNEL);
- if (!skb)
+ if (!skb) {
+ struct sock *sk;
+
+ sk = netlink_lookup(in_skb->sk->sk_protocol,
+ NETLINK_CB(in_skb).pid);
+ if (sk) {
+ sk->sk_err = ENOBUFS;
+ sk->sk_error_report(sk);
+ }
return;
+ }
rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
NLMSG_ERROR, sizeof(struct nlmsgerr));
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
