[AF_KEY]: Dump SA/SP entries non-atomically

Stop dumping of entries when af_key socket receive queue is getting
full and continue it later when there is more room again.

This fixes dumping of large databases. Currently the entries not
fitting into the receive queue are just dropped (including the
end-of-dump message) which can confuse applications.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/key/af_key.c

@@ -48,6 +48,17 @@ struct pfkey_sock {
 	struct sock	sk;
 	int		registered;
 	int		promisc;
+
+	struct {
+		uint8_t		msg_version;
+		uint32_t	msg_pid;
+		int		(*dump)(struct pfkey_sock *sk);
+		void		(*done)(struct pfkey_sock *sk);
+		union {
+			struct xfrm_policy_walk	policy;
+			struct xfrm_state_walk	state;
+		} u;
+	} dump;
 };
 
 static inline struct pfkey_sock *pfkey_sk(struct sock *sk)
@@ -55,6 +66,27 @@ static inline struct pfkey_sock *pfkey_sk(struct sock *sk)
 	return (struct pfkey_sock *)sk;
 }
 
+static int pfkey_can_dump(struct sock *sk)
+{
+	if (3 * atomic_read(&sk->sk_rmem_alloc) <= 2 * sk->sk_rcvbuf)
+		return 1;
+	return 0;
+}
+
+static int pfkey_do_dump(struct pfkey_sock *pfk)
+{
+	int rc;
+
+	rc = pfk->dump.dump(pfk);
+	if (rc == -ENOBUFS)
+		return 0;
+
+	pfk->dump.done(pfk);
+	pfk->dump.dump = NULL;
+	pfk->dump.done = NULL;
+	return rc;
+}
+
 static void pfkey_sock_destruct(struct sock *sk)
 {
 	skb_queue_purge(&sk->sk_receive_queue);
@@ -1709,51 +1741,60 @@ static int pfkey_flush(struct sock *sk, struct sk_buff *skb, struct sadb_msg *hd
 	return 0;
 }
 
-struct pfkey_dump_data
-{
-	struct sk_buff *skb;
-	struct sadb_msg *hdr;
-	struct sock *sk;
-};
-
 static int dump_sa(struct xfrm_state *x, int count, void *ptr)
 {
-	struct pfkey_dump_data *data = ptr;
+	struct pfkey_sock *pfk = ptr;
 	struct sk_buff *out_skb;
 	struct sadb_msg *out_hdr;
 
+	if (!pfkey_can_dump(&pfk->sk))
+		return -ENOBUFS;
+
 	out_skb = pfkey_xfrm_state2msg(x);
 	if (IS_ERR(out_skb))
 		return PTR_ERR(out_skb);
 
 	out_hdr = (struct sadb_msg *) out_skb->data;
-	out_hdr->sadb_msg_version = data->hdr->sadb_msg_version;
+	out_hdr->sadb_msg_version = pfk->dump.msg_version;
 	out_hdr->sadb_msg_type = SADB_DUMP;
 	out_hdr->sadb_msg_satype = pfkey_proto2satype(x->id.proto);
 	out_hdr->sadb_msg_errno = 0;
 	out_hdr->sadb_msg_reserved = 0;
 	out_hdr->sadb_msg_seq = count;
-	out_hdr->sadb_msg_pid = data->hdr->sadb_msg_pid;
-	pfkey_broadcast(out_skb, GFP_ATOMIC, BROADCAST_ONE, data->sk);
+	out_hdr->sadb_msg_pid = pfk->dump.msg_pid;
+	pfkey_broadcast(out_skb, GFP_ATOMIC, BROADCAST_ONE, &pfk->sk);
 	return 0;
 }
 
+static int pfkey_dump_sa(struct pfkey_sock *pfk)
+{
+	return xfrm_state_walk(&pfk->dump.u.state, dump_sa, (void *) pfk);
+}
+
+static void pfkey_dump_sa_done(struct pfkey_sock *pfk)
+{
+	xfrm_state_walk_done(&pfk->dump.u.state);
+}
+
 static int pfkey_dump(struct sock *sk, struct sk_buff *skb, struct sadb_msg *hdr, void **ext_hdrs)
 {
 	u8 proto;
-	struct pfkey_dump_data data = { .skb = skb, .hdr = hdr, .sk = sk };
-	struct xfrm_state_walk walk;
-	int rc;
+	struct pfkey_sock *pfk = pfkey_sk(sk);
+
+	if (pfk->dump.dump != NULL)
+		return -EBUSY;
 
 	proto = pfkey_satype2proto(hdr->sadb_msg_satype);
 	if (proto == 0)
 		return -EINVAL;
 
-	xfrm_state_walk_init(&walk, proto);
-	rc = xfrm_state_walk(&walk, dump_sa, &data);
-	xfrm_state_walk_done(&walk);
+	pfk->dump.msg_version = hdr->sadb_msg_version;
+	pfk->dump.msg_pid = hdr->sadb_msg_pid;
+	pfk->dump.dump = pfkey_dump_sa;
+	pfk->dump.done = pfkey_dump_sa_done;
+	xfrm_state_walk_init(&pfk->dump.u.state, proto);
 
-	return rc;
+	return pfkey_do_dump(pfk);
 }
 
 static int pfkey_promisc(struct sock *sk, struct sk_buff *skb, struct sadb_msg *hdr, void **ext_hdrs)
@@ -2648,11 +2689,14 @@ out:
 
 static int dump_sp(struct xfrm_policy *xp, int dir, int count, void *ptr)
 {
-	struct pfkey_dump_data *data = ptr;
+	struct pfkey_sock *pfk = ptr;
 	struct sk_buff *out_skb;
 	struct sadb_msg *out_hdr;
 	int err;
 
+	if (!pfkey_can_dump(&pfk->sk))
+		return -ENOBUFS;
+
 	out_skb = pfkey_xfrm_policy2msg_prep(xp);
 	if (IS_ERR(out_skb))
 		return PTR_ERR(out_skb);
@@ -2662,27 +2706,40 @@ static int dump_sp(struct xfrm_policy *xp, int dir, int count, void *ptr)
 		return err;
 
 	out_hdr = (struct sadb_msg *) out_skb->data;
-	out_hdr->sadb_msg_version = data->hdr->sadb_msg_version;
+	out_hdr->sadb_msg_version = pfk->dump.msg_version;
 	out_hdr->sadb_msg_type = SADB_X_SPDDUMP;
 	out_hdr->sadb_msg_satype = SADB_SATYPE_UNSPEC;
 	out_hdr->sadb_msg_errno = 0;
 	out_hdr->sadb_msg_seq = count;
-	out_hdr->sadb_msg_pid = data->hdr->sadb_msg_pid;
-	pfkey_broadcast(out_skb, GFP_ATOMIC, BROADCAST_ONE, data->sk);
+	out_hdr->sadb_msg_pid = pfk->dump.msg_pid;
+	pfkey_broadcast(out_skb, GFP_ATOMIC, BROADCAST_ONE, &pfk->sk);
 	return 0;
 }
 
+static int pfkey_dump_sp(struct pfkey_sock *pfk)
+{
+	return xfrm_policy_walk(&pfk->dump.u.policy, dump_sp, (void *) pfk);
+}
+
+static void pfkey_dump_sp_done(struct pfkey_sock *pfk)
+{
+	xfrm_policy_walk_done(&pfk->dump.u.policy);
+}
+
 static int pfkey_spddump(struct sock *sk, struct sk_buff *skb, struct sadb_msg *hdr, void **ext_hdrs)
 {
-	struct pfkey_dump_data data = { .skb = skb, .hdr = hdr, .sk = sk };
-	struct xfrm_policy_walk walk;
-	int rc;
+	struct pfkey_sock *pfk = pfkey_sk(sk);
 
-	xfrm_policy_walk_init(&walk, XFRM_POLICY_TYPE_MAIN);
-	rc = xfrm_policy_walk(&walk, dump_sp, &data);
-	xfrm_policy_walk_done(&walk);
+	if (pfk->dump.dump != NULL)
+		return -EBUSY;
 
-	return rc;
+	pfk->dump.msg_version = hdr->sadb_msg_version;
+	pfk->dump.msg_pid = hdr->sadb_msg_pid;
+	pfk->dump.dump = pfkey_dump_sp;
+	pfk->dump.done = pfkey_dump_sp_done;
+	xfrm_policy_walk_init(&pfk->dump.u.policy, XFRM_POLICY_TYPE_MAIN);
+
+	return pfkey_do_dump(pfk);
 }
 
 static int key_notify_policy_flush(struct km_event *c)
@@ -3687,6 +3744,7 @@ static int pfkey_recvmsg(struct kiocb *kiocb,
 			 int flags)
 {
 	struct sock *sk = sock->sk;
+	struct pfkey_sock *pfk = pfkey_sk(sk);
 	struct sk_buff *skb;
 	int copied, err;
 
@@ -3714,6 +3772,10 @@ static int pfkey_recvmsg(struct kiocb *kiocb,
 
 	err = (flags & MSG_TRUNC) ? skb->len : copied;
 
+	if (pfk->dump.dump != NULL &&
+	    3 * atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)
+		pfkey_do_dump(pfk);
+
 out_free:
 	skb_free_datagram(sk, skb);
 out: