@@ -46,6 +46,7 @@ static unsigned int nfnetlink_pernet_id __read_mostly;
struct nfnl_net {
struct sock *nfnl;
+ struct sock *nfnl_stash;
};
static struct {
@@ -160,37 +161,56 @@ nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
return &ss->cb[cb_id];
}
-int nfnetlink_has_listeners(struct net *net, unsigned int group)
+static struct sock *nfnl_pernet_sk(struct net *net)
{
struct nfnl_net *nfnlnet = nfnl_pernet(net);
- return netlink_has_listeners(nfnlnet->nfnl, group);
+ return READ_ONCE(nfnlnet->nfnl);
+}
+
+int nfnetlink_has_listeners(struct net *net, unsigned int group)
+{
+ struct sock *nlsk = nfnl_pernet_sk(net);
+
+ return nlsk ? netlink_has_listeners(nlsk, group) : 0;
}
EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
unsigned int group, int echo, gfp_t flags)
{
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
+
+ if (nlsk)
+ return nlmsg_notify(nlsk, skb, portid, group, echo, flags);
- return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
+ /* nlsk already gone? This happens when .pre_exit was already called,
+ * return 0, we can't retry.
+ */
+ kfree_skb(skb);
+ return 0;
}
EXPORT_SYMBOL_GPL(nfnetlink_send);
int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
{
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
- return netlink_set_err(nfnlnet->nfnl, portid, group, error);
+ return nlsk ? netlink_set_err(nlsk, portid, group, error) : 0;
}
EXPORT_SYMBOL_GPL(nfnetlink_set_err);
int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
{
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
int err;
- err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
+ if (!nlsk) {
+ kfree_skb(skb);
+ return 0;
+ }
+
+ err = nlmsg_unicast(nlsk, skb, portid);
if (err == -EAGAIN)
err = -ENOBUFS;
@@ -201,9 +221,12 @@ EXPORT_SYMBOL_GPL(nfnetlink_unicast);
void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
__u32 group, gfp_t allocation)
{
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
- netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
+ if (nlsk)
+ netlink_broadcast(nlsk, skb, portid, group, allocation);
+ else
+ kfree_skb(skb);
}
EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
@@ -247,7 +270,7 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
{
int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
struct nlattr *attr = (void *)nlh + min_len;
@@ -255,7 +278,7 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
__u8 subsys_id = NFNL_SUBSYS_ID(type);
struct nfnl_info info = {
.net = net,
- .sk = nfnlnet->nfnl,
+ .sk = nlsk,
.nlh = nlh,
.nfmsg = nlmsg_data(nlh),
.extack = extack,
@@ -484,14 +507,14 @@ static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
{
int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
- struct nfnl_net *nfnlnet = nfnl_pernet(net);
+ struct sock *nlsk = nfnl_pernet_sk(net);
struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
struct nlattr *attr = (void *)nlh + min_len;
u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
int attrlen = nlh->nlmsg_len - min_len;
struct nfnl_info info = {
.net = net,
- .sk = nfnlnet->nfnl,
+ .sk = nlsk,
.nlh = nlh,
.nfmsg = nlmsg_data(nlh),
.extack = &extack,
@@ -699,12 +722,21 @@ static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
list_for_each_entry(net, net_exit_list, exit_list) {
nfnlnet = nfnl_pernet(net);
- netlink_kernel_release(nfnlnet->nfnl);
+ netlink_kernel_release(nfnlnet->nfnl_stash);
}
}
+static void __net_exit nfnetlink_net_pre_exit(struct net *net)
+{
+ struct nfnl_net *nfnlnet = nfnl_pernet(net);
+
+ nfnlnet->nfnl_stash = nfnlnet->nfnl;
+ WRITE_ONCE(nfnlnet->nfnl, NULL);
+}
+
static struct pernet_operations nfnetlink_net_ops = {
.init = nfnetlink_net_init,
+ .pre_exit = nfnetlink_net_pre_exit,
.exit_batch = nfnetlink_net_exit_batch,
.id = &nfnetlink_pernet_id,
.size = sizeof(struct nfnl_net),
At this time upper layer, e.g. nf_conntrack_event, has to make sure that its pernet exit handler runs before the nfnetlink one, otherwise we get a crash if kernel tries to send a conntrack event after the nfnetlink netns exit handler did close the socket already. In order to move nf_conntrack_ecache to global (not pernet) netns event pointer again the nfnetlink apis need to survive attempts to send a netlink message after the socket has been destroyed in nfnetlink netns exit function. Set the pernet socket to null in the pre_exit handler and close it in the exit_batch handler via a 'stash' pointer. All functions now check nlsk for NULL before using it. Signed-off-by: Florian Westphal <fw@strlen.de> --- net/netfilter/nfnetlink.c | 62 +++++++++++++++++++++++++++++---------- 1 file changed, 47 insertions(+), 15 deletions(-)