]> git.itanic.dy.fi Git - linux-stable/commitdiff
wireguard: netlink: check for dangling peer via is_dead instead of empty list
authorJason A. Donenfeld <Jason@zx2c4.com>
Thu, 14 Mar 2024 22:49:09 +0000 (16:49 -0600)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 3 Apr 2024 13:32:26 +0000 (15:32 +0200)
[ Upstream commit 55b6c738673871c9b0edae05d0c97995c1ff08c4 ]

If all peers are removed via wg_peer_remove_all(), rather than setting
peer_list to empty, the peer is added to a temporary list with a head on
the stack of wg_peer_remove_all(). If a netlink dump is resumed and the
cursored peer is one that has been removed via wg_peer_remove_all(), it
will iterate from that peer and then attempt to dump freed peers.

Fix this by instead checking peer->is_dead, which was explictly created
for this purpose. Also move up the device_update_lock lockdep assertion,
since reading is_dead relies on that.

It can be reproduced by a small script like:

    echo "Setting config..."
    ip link add dev wg0 type wireguard
    wg setconf wg0 /big-config
    (
            while true; do
                    echo "Showing config..."
                    wg showconf wg0 > /dev/null
            done
    ) &
    sleep 4
    wg setconf wg0 <(printf "[Peer]\nPublicKey=$(wg genkey)\n")

Resulting in:

    BUG: KASAN: slab-use-after-free in __lock_acquire+0x182a/0x1b20
    Read of size 8 at addr ffff88811956ec70 by task wg/59
    CPU: 2 PID: 59 Comm: wg Not tainted 6.8.0-rc2-debug+ #5
    Call Trace:
     <TASK>
     dump_stack_lvl+0x47/0x70
     print_address_description.constprop.0+0x2c/0x380
     print_report+0xab/0x250
     kasan_report+0xba/0xf0
     __lock_acquire+0x182a/0x1b20
     lock_acquire+0x191/0x4b0
     down_read+0x80/0x440
     get_peer+0x140/0xcb0
     wg_get_device_dump+0x471/0x1130

Cc: stable@vger.kernel.org
Fixes: e7096c131e51 ("net: WireGuard secure network tunnel")
Reported-by: Lillian Berry <lillian@star-ark.net>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/net/wireguard/netlink.c

index e220d761b1f27aa31eab3ad1b9211ddfe55eaabd..c17aee454fa3b4ec9585126ccd7b8c0f3290622c 100644 (file)
@@ -255,17 +255,17 @@ static int wg_get_device_dump(struct sk_buff *skb, struct netlink_callback *cb)
        if (!peers_nest)
                goto out;
        ret = 0;
-       /* If the last cursor was removed via list_del_init in peer_remove, then
+       lockdep_assert_held(&wg->device_update_lock);
+       /* If the last cursor was removed in peer_remove or peer_remove_all, then
         * we just treat this the same as there being no more peers left. The
         * reason is that seq_nr should indicate to userspace that this isn't a
         * coherent dump anyway, so they'll try again.
         */
        if (list_empty(&wg->peer_list) ||
-           (ctx->next_peer && list_empty(&ctx->next_peer->peer_list))) {
+           (ctx->next_peer && ctx->next_peer->is_dead)) {
                nla_nest_cancel(skb, peers_nest);
                goto out;
        }
-       lockdep_assert_held(&wg->device_update_lock);
        peer = list_prepare_entry(ctx->next_peer, &wg->peer_list, peer_list);
        list_for_each_entry_continue(peer, &wg->peer_list, peer_list) {
                if (get_peer(peer, skb, ctx)) {