Skip to contentSkip to conversation

netfilter: ctnetlink: reject expectation deletions on unconfirmed conntrack

PATCH In review Pablo Neira Ayuso 1 participant 3 messages Jul 29

Subsystems (2)
lore
Share & tools
lore

Pablo Neira Ayuso

author maintainer
#0 lore raw

PATCH v2 1/2 netfilter: ctnetlink: reject expectation deletions on unconfirmed conntrack

ctnetlink allows to reach expectations that refer to unconfirmed conntrack. The ct extension area of unconfirmed conntrack could be reallocated while the packet is in transit.

While holding the spinlock, check if the master conntrack is unconfirmed. The existing expectation spinlock ensures that the master conntrack is still alive because nf_ct_remove_expectations() tears down all expectations on conntrack removal. So, while holding the spinlock, the lifetime of the expectations is attached to the master conntrack.

Fixes ecfab2c9fe55 [NETFILTER]: nf_conntrack: introduce extension infrastructure

---
v2: new in this series, to address sashiko report.

 net/netfilter/nf_conntrack_expect.c  | 3 +++
 net/netfilter/nf_conntrack_netlink.c | 6 ++++++
 2 files changed, 9 insertions(+)
net/netfilter/nf_conntrack_expect.c +3 -0
index 8a3b9e33e94f..7694deaa5b12 100644
@@ -620,6 +620,9 @@ void nf_ct_expect_iterate_net(struct net *net,
 			if (!net_eq(nf_ct_exp_net(exp), net))
 				continue;
 
+			if (!nf_ct_is_confirmed(exp->master))
+				continue;
+
 			if (iter(exp, data))
 				nf_ct_unlink_expect_report(exp, portid, report);
 		}
net/netfilter/nf_conntrack_netlink.c +6 -0
index 31cbb1b55b9e..b58f51372849 100644
@@ -3442,6 +3442,12 @@ static int ctnetlink_del_expect(struct sk_buff *skb,
 			}
 		}
 
+		if (!nf_ct_is_confirmed(exp->master)) {
+			nf_ct_expect_put(exp);
+			spin_unlock_bh(&nf_conntrack_expect_lock);
+			return -EBUSY;
+		}
+
 		/* after list removal, usage count == 1 */
 		nf_ct_unlink_expect_report(exp, NETLINK_CB(skb).portid,
 					   nlmsg_report(info->nlh));
Signature · 1 line
2.47.3
author maintainer
#1 lore raw

PATCH v2 2/2 netfilter: nf_conntrack: move expectation hlist_head on ct extension realloc

Commit 7c9664351980 ("netfilter: move nat hlist_head to nf_conn") moved the nat hlist_head to nf_conn and it removes the .move callback for ct extensions. This commit description already points to the same problem that is being addressed for expectations by this patch:

1. ...
2. When reallocation of extension area occurs we need to fixup the bysource hash head via hlist_replace_rcu.

This means that the .pprev pointer of the first expectation might refer to a stale hlist_head after reallocation.

This bug is now easier to trigger since the introduction of the commit 857b46027d6f ("netfilter: nft_ct: add ct expectations support") which allows to create expectations before the ct extension area have been fully set up for unconfirmed conntracks.

This patch uses hlist_move_list() because this conntrack is unconfirmed, ie. not yet in the hashes, and it is neither visible to the conntrack garbage collector nor ctnetlink.

Fixes 857b46027d6f ("netfilter: nft_ct: add ct expectations support")

---
v2: no changes.

 net/netfilter/nf_conntrack_extend.c | 33 ++++++++++++++++++++++++++++-
 1 file changed, 32 insertions(+), 1 deletion(-)
net/netfilter/nf_conntrack_extend.c +32 -1
index 0da105e1ded9..ea7795a2461f 100644
@@ -87,9 +87,34 @@ static __always_inline unsigned int total_extension_size(void)
 	;
 }
 
+static void nf_ct_ext_helper_save(struct nf_conn *ct,
+				  struct hlist_head *expectations)
+{
+	struct nf_conn_help *help;
+
+	help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
+	if (!help)
+		return;
+
+	hlist_move_list(&help->expectations, expectations);
+}
+
+static void nf_ct_ext_helper_restore(struct nf_conn *ct,
+				     struct hlist_head *expectations)
+{
+	struct nf_conn_help *help;
+
+	help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
+	if (!help)
+		return;
+
+	hlist_move_list(expectations, &help->expectations);
+}
+
 void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 {
 	unsigned int newlen, newoff, oldlen, alloc;
+	HLIST_HEAD(expectations);
 	struct nf_ct_ext *new;
 
 	/* Conntrack must not be confirmed to avoid races on reallocation. */
@@ -108,13 +133,17 @@ void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 		oldlen = sizeof(*new);
 	}
 
+	nf_ct_ext_helper_save(ct, &expectations);
+
 	newoff = ALIGN(oldlen, __alignof__(struct nf_ct_ext));
 	newlen = newoff + nf_ct_ext_type_len[id];
 
 	alloc = max(newlen, NF_CT_EXT_PREALLOC);
 	new = krealloc(ct->ext, alloc, gfp);
-	if (!new)
+	if (!new) {
+		nf_ct_ext_helper_restore(ct, &expectations);
 		return NULL;
+	}
 
 	if (!ct->ext)
 		memset(new->offset, 0, sizeof(new->offset));
@@ -124,6 +153,8 @@ void *nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 	memset((void *)new + newoff, 0, newlen - newoff);
 
 	ct->ext = new;
+	nf_ct_ext_helper_restore(ct, &expectations);
+
 	return (void *)new + newoff;
 }
 EXPORT_SYMBOL(nf_ct_ext_add);
Signature · 1 line
2.47.3
author maintainer
#2 lore raw
In reply to Pablo Neira Ayuso
Quoted 9 lines · Pablo Neira Ayuso

ctnetlink allows to reach expectations that refer to unconfirmed conntrack. The ct extension area of unconfirmed conntrack could be reallocated while the packet is in transit.

While holding the spinlock, check if the master conntrack is unconfirmed. The existing expectation spinlock ensures that the master conntrack is still alive because nf_ct_remove_expectations() tears down all expectations on conntrack removal. So, while holding the spinlock, the lifetime of the expectations is attached to the master conntrack.

Scratch this series, I'm not tackling the issue from the right angle.

Will follow up asap.