bpf: Add link-based BPF program attachment to network namespace

Extend bpf() syscall subcommands that operate on bpf_link, that is
LINK_CREATE, LINK_UPDATE, OBJ_GET_INFO, to accept attach types tied to
network namespaces (only flow dissector at the moment).

Link-based and prog-based attachment can be used interchangeably, but only
one can exist at a time. Attempts to attach a link when a prog is already
attached directly, and the other way around, will be met with -EEXIST.
Attempts to detach a program when link exists result in -EINVAL.

Attachment of multiple links of same attach type to one netns is not
supported with the intention to lift the restriction when a use-case
presents itself. Because of that link create returns -E2BIG when trying to
create another netns link, when one already exists.

Link-based attachments to netns don't keep a netns alive by holding a ref
to it. Instead links get auto-detached from netns when the latter is being
destroyed, using a pernet pre_exit callback.

When auto-detached, link lives in defunct state as long there are open FDs
for it. -ENOLINK is returned if a user tries to update a defunct link.

Because bpf_link to netns doesn't hold a ref to struct net, special care is
taken when releasing, updating, or filling link info. The netns might be
getting torn down when any of these link operations are in progress. That
is why auto-detach and update/release/fill_info are synchronized by the
same mutex. Also, link ops have to always check if auto-detach has not
happened yet and if netns is still alive (refcnt > 0).

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-5-jakub@cloudflare.com

2 files changed, 245 insertions, 2 deletions

diff --git a/kernel/bpf/net_namespace.c b/kernel/bpf/net_namespace.c
index b37d81450c3a..78cf061f8179 100644
--- a/kernel/bpf/net_namespace.c
+++ b/kernel/bpf/net_namespace.c
@@ -8,9 +8,140 @@
  * Functions to manage BPF programs attached to netns
  */
 
+struct bpf_netns_link {
+	struct bpf_link	link;
+	enum bpf_attach_type type;
+	enum netns_bpf_attach_type netns_type;
+
+	/* We don't hold a ref to net in order to auto-detach the link
+	 * when netns is going away. Instead we rely on pernet
+	 * pre_exit callback to clear this pointer. Must be accessed
+	 * with netns_bpf_mutex held.
+	 */
+	struct net *net;
+};
+
 /* Protects updates to netns_bpf */
 DEFINE_MUTEX(netns_bpf_mutex);
 
+/* Must be called with netns_bpf_mutex held. */
+static void __net_exit bpf_netns_link_auto_detach(struct bpf_link *link)
+{
+	struct bpf_netns_link *net_link =
+		container_of(link, struct bpf_netns_link, link);
+
+	net_link->net = NULL;
+}
+
+static void bpf_netns_link_release(struct bpf_link *link)
+{
+	struct bpf_netns_link *net_link =
+		container_of(link, struct bpf_netns_link, link);
+	enum netns_bpf_attach_type type = net_link->netns_type;
+	struct net *net;
+
+	/* Link auto-detached by dying netns. */
+	if (!net_link->net)
+		return;
+
+	mutex_lock(&netns_bpf_mutex);
+
+	/* Recheck after potential sleep. We can race with cleanup_net
+	 * here, but if we see a non-NULL struct net pointer pre_exit
+	 * has not happened yet and will block on netns_bpf_mutex.
+	 */
+	net = net_link->net;
+	if (!net)
+		goto out_unlock;
+
+	net->bpf.links[type] = NULL;
+	RCU_INIT_POINTER(net->bpf.progs[type], NULL);
+
+out_unlock:
+	mutex_unlock(&netns_bpf_mutex);
+}
+
+static void bpf_netns_link_dealloc(struct bpf_link *link)
+{
+	struct bpf_netns_link *net_link =
+		container_of(link, struct bpf_netns_link, link);
+
+	kfree(net_link);
+}
+
+static int bpf_netns_link_update_prog(struct bpf_link *link,
+				      struct bpf_prog *new_prog,
+				      struct bpf_prog *old_prog)
+{
+	struct bpf_netns_link *net_link =
+		container_of(link, struct bpf_netns_link, link);
+	enum netns_bpf_attach_type type = net_link->netns_type;
+	struct net *net;
+	int ret = 0;
+
+	if (old_prog && old_prog != link->prog)
+		return -EPERM;
+	if (new_prog->type != link->prog->type)
+		return -EINVAL;
+
+	mutex_lock(&netns_bpf_mutex);
+
+	net = net_link->net;
+	if (!net || !check_net(net)) {
+		/* Link auto-detached or netns dying */
+		ret = -ENOLINK;
+		goto out_unlock;
+	}
+
+	old_prog = xchg(&link->prog, new_prog);
+	rcu_assign_pointer(net->bpf.progs[type], new_prog);
+	bpf_prog_put(old_prog);
+
+out_unlock:
+	mutex_unlock(&netns_bpf_mutex);
+	return ret;
+}
+
+static int bpf_netns_link_fill_info(const struct bpf_link *link,
+				    struct bpf_link_info *info)
+{
+	const struct bpf_netns_link *net_link =
+		container_of(link, struct bpf_netns_link, link);
+	unsigned int inum = 0;
+	struct net *net;
+
+	mutex_lock(&netns_bpf_mutex);
+	net = net_link->net;
+	if (net && check_net(net))
+		inum = net->ns.inum;
+	mutex_unlock(&netns_bpf_mutex);
+
+	info->netns.netns_ino = inum;
+	info->netns.attach_type = net_link->type;
+	return 0;
+}
+
+static void bpf_netns_link_show_fdinfo(const struct bpf_link *link,
+				       struct seq_file *seq)
+{
+	struct bpf_link_info info = {};
+
+	bpf_netns_link_fill_info(link, &info);
+	seq_printf(seq,
+		   "netns_ino:\t%u\n"
+		   "attach_type:\t%u\n",
+		   info.netns.netns_ino,
+		   info.netns.attach_type);
+}
+
+static const struct bpf_link_ops bpf_netns_link_ops = {
+	.release = bpf_netns_link_release,
+	.dealloc = bpf_netns_link_dealloc,
+	.update_prog = bpf_netns_link_update_prog,
+	.fill_link_info = bpf_netns_link_fill_info,
+	.show_fdinfo = bpf_netns_link_show_fdinfo,
+};
+
 int netns_bpf_prog_query(const union bpf_attr *attr,
 			 union bpf_attr __user *uattr)
 {
@@ -67,6 +198,13 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 
 	net = current->nsproxy->net_ns;
 	mutex_lock(&netns_bpf_mutex);
+
+	/* Attaching prog directly is not compatible with links */
+	if (net->bpf.links[type]) {
+		ret = -EEXIST;
+		goto out_unlock;
+	}
+
 	switch (type) {
 	case NETNS_BPF_FLOW_DISSECTOR:
 		ret = flow_dissector_bpf_prog_attach(net, prog);
@@ -75,6 +213,7 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 		ret = -EINVAL;
 		break;
 	}
+out_unlock:
 	mutex_unlock(&netns_bpf_mutex);
 
 	return ret;
@@ -86,6 +225,10 @@ static int __netns_bpf_prog_detach(struct net *net,
 {
 	struct bpf_prog *attached;
 
+	/* Progs attached via links cannot be detached */
+	if (net->bpf.links[type])
+		return -EINVAL;
+
 	attached = rcu_dereference_protected(net->bpf.progs[type],
 					     lockdep_is_held(&netns_bpf_mutex));
 	if (!attached)
@@ -111,13 +254,110 @@ int netns_bpf_prog_detach(const union bpf_attr *attr)
 	return ret;
 }
 
+static int netns_bpf_link_attach(struct net *net, struct bpf_link *link,
+				 enum netns_bpf_attach_type type)
+{
+	struct bpf_prog *prog;
+	int err;
+
+	mutex_lock(&netns_bpf_mutex);
+
+	/* Allow attaching only one prog or link for now */
+	if (net->bpf.links[type]) {
+		err = -E2BIG;
+		goto out_unlock;
+	}
+	/* Links are not compatible with attaching prog directly */
+	prog = rcu_dereference_protected(net->bpf.progs[type],
+					 lockdep_is_held(&netns_bpf_mutex));
+	if (prog) {
+		err = -EEXIST;
+		goto out_unlock;
+	}
+
+	switch (type) {
+	case NETNS_BPF_FLOW_DISSECTOR:
+		err = flow_dissector_bpf_prog_attach(net, link->prog);
+		break;
+	default:
+		err = -EINVAL;
+		break;
+	}
+	if (err)
+		goto out_unlock;
+
+	net->bpf.links[type] = link;
+
+out_unlock:
+	mutex_unlock(&netns_bpf_mutex);
+	return err;
+}
+
+int netns_bpf_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
+{
+	enum netns_bpf_attach_type netns_type;
+	struct bpf_link_primer link_primer;
+	struct bpf_netns_link *net_link;
+	enum bpf_attach_type type;
+	struct net *net;
+	int err;
+
+	if (attr->link_create.flags)
+		return -EINVAL;
+
+	type = attr->link_create.attach_type;
+	netns_type = to_netns_bpf_attach_type(type);
+	if (netns_type < 0)
+		return -EINVAL;
+
+	net = get_net_ns_by_fd(attr->link_create.target_fd);
+	if (IS_ERR(net))
+		return PTR_ERR(net);
+
+	net_link = kzalloc(sizeof(*net_link), GFP_USER);
+	if (!net_link) {
+		err = -ENOMEM;
+		goto out_put_net;
+	}
+	bpf_link_init(&net_link->link, BPF_LINK_TYPE_NETNS,
+		      &bpf_netns_link_ops, prog);
+	net_link->net = net;
+	net_link->type = type;
+	net_link->netns_type = netns_type;
+
+	err = bpf_link_prime(&net_link->link, &link_primer);
+	if (err) {
+		kfree(net_link);
+		goto out_put_net;
+	}
+
+	err = netns_bpf_link_attach(net, &net_link->link, netns_type);
+	if (err) {
+		bpf_link_cleanup(&link_primer);
+		goto out_put_net;
+	}
+
+	put_net(net);
+	return bpf_link_settle(&link_primer);
+
+out_put_net:
+	put_net(net);
+	return err;
+}
+
 static void __net_exit netns_bpf_pernet_pre_exit(struct net *net)
 {
 	enum netns_bpf_attach_type type;
+	struct bpf_link *link;
 
 	mutex_lock(&netns_bpf_mutex);
-	for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++)
-		__netns_bpf_prog_detach(net, type);
+	for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++) {
+		link = net->bpf.links[type];
+		if (link)
+			bpf_netns_link_auto_detach(link);
+		else
+			__netns_bpf_prog_detach(net, type);
+	}
 	mutex_unlock(&netns_bpf_mutex);
 }
 
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index c77ab9c76f7b..e14a842d7e0d 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -3887,6 +3887,9 @@ static int link_create(union bpf_attr *attr)
 	case BPF_PROG_TYPE_TRACING:
 		ret = tracing_bpf_link_attach(attr, prog);
 		break;
+	case BPF_PROG_TYPE_FLOW_DISSECTOR:
+		ret = netns_bpf_link_create(attr, prog);
+		break;
 	default:
 		ret = -EINVAL;
 	}