]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commitdiff
ipv6: addrconf: don't use rtnl mutex in RTM_GETNETCONF
authorFlorian Westphal <fw@strlen.de>
Wed, 11 Oct 2017 08:28:00 +0000 (10:28 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 12 Oct 2017 03:17:03 +0000 (20:17 -0700)
Instead of relying on rtnl mutex bump device reference count.
After this change, values reported can change in parallel, but thats not
much different from current state, as anyone can change the settings
right after rtnl_unlock (and before userspace processed reply).

While at it, switch to GFP_KERNEL allocation.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/addrconf.c

index d9f6226694ebaf4beac01684e3fba7b35e0d81de..5207f567ef28c777d02f91307c63581ae75a3f5e 100644 (file)
@@ -616,23 +616,23 @@ static int inet6_netconf_get_devconf(struct sk_buff *in_skb,
 {
        struct net *net = sock_net(in_skb->sk);
        struct nlattr *tb[NETCONFA_MAX+1];
+       struct inet6_dev *in6_dev = NULL;
+       struct net_device *dev = NULL;
        struct netconfmsg *ncm;
        struct sk_buff *skb;
        struct ipv6_devconf *devconf;
-       struct inet6_dev *in6_dev;
-       struct net_device *dev;
        int ifindex;
        int err;
 
        err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
                          devconf_ipv6_policy, extack);
        if (err < 0)
-               goto errout;
+               return err;
 
-       err = -EINVAL;
        if (!tb[NETCONFA_IFINDEX])
-               goto errout;
+               return -EINVAL;
 
+       err = -EINVAL;
        ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
        switch (ifindex) {
        case NETCONFA_IFINDEX_ALL:
@@ -642,10 +642,10 @@ static int inet6_netconf_get_devconf(struct sk_buff *in_skb,
                devconf = net->ipv6.devconf_dflt;
                break;
        default:
-               dev = __dev_get_by_index(net, ifindex);
+               dev = dev_get_by_index(net, ifindex);
                if (!dev)
-                       goto errout;
-               in6_dev = __in6_dev_get(dev);
+                       return -EINVAL;
+               in6_dev = in6_dev_get(dev);
                if (!in6_dev)
                        goto errout;
                devconf = &in6_dev->cnf;
@@ -653,7 +653,7 @@ static int inet6_netconf_get_devconf(struct sk_buff *in_skb,
        }
 
        err = -ENOBUFS;
-       skb = nlmsg_new(inet6_netconf_msgsize_devconf(NETCONFA_ALL), GFP_ATOMIC);
+       skb = nlmsg_new(inet6_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
        if (!skb)
                goto errout;
 
@@ -669,6 +669,10 @@ static int inet6_netconf_get_devconf(struct sk_buff *in_skb,
        }
        err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
 errout:
+       if (in6_dev)
+               in6_dev_put(in6_dev);
+       if (dev)
+               dev_put(dev);
        return err;
 }
 
@@ -6570,7 +6574,7 @@ int __init addrconf_init(void)
        __rtnl_register(PF_INET6, RTM_GETANYCAST, NULL,
                        inet6_dump_ifacaddr, 0);
        __rtnl_register(PF_INET6, RTM_GETNETCONF, inet6_netconf_get_devconf,
-                       inet6_netconf_dump_devconf, 0);
+                       inet6_netconf_dump_devconf, RTNL_FLAG_DOIT_UNLOCKED);
 
        ipv6_addr_label_rtnl_register();