net: Only allow init netns to set default tcp cong to a restricted algo

commit 8d43259 upstream.

tcp_set_default_congestion_control() is netns-safe in that it writes
to &net->ipv4.tcp_congestion_control, but it also sets
ca->flags |= TCP_CONG_NON_RESTRICTED which is not namespaced.
This has the unintended side-effect of changing the global
net.ipv4.tcp_allowed_congestion_control sysctl, despite the fact that it
is read-only: 97684f0 ("net: Make tcp_allowed_congestion_control
readonly in non-init netns")

Resolve this netns "leak" by only allowing the init netns to set the
default algorithm to one that is restricted. This restriction could be
removed if tcp_allowed_congestion_control were namespace-ified in the
future.

This bug was uncovered with
https://github.com/JonathonReinhart/linux-netns-sysctl-verify

Fixes: 6670e15 ("tcp: Namespace-ify sysctl_tcp_default_congestion_control")
Signed-off-by: Jonathon Reinhart <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: JonathonReinhart

net/ipv4/tcp_cong.c

@@ -229,6 +229,10 @@ int tcp_set_default_congestion_control(struct net *net, const char *name)
 		ret = -ENOENT;
 	} else if (!try_module_get(ca->owner)) {
 		ret = -EBUSY;
+	} else if (!net_eq(net, &init_net) &&
+			!(ca->flags & TCP_CONG_NON_RESTRICTED)) {
+		/* Only init netns can set default to a restricted algorithm */
+		ret = -EPERM;
 	} else {
 		prev = xchg(&net->ipv4.tcp_congestion_control, ca);
 		if (prev)