[PATCH net 0/3] Netfilter fixes for net

[PATCH net 0/3] Netfilter fixes for net

[Pablo Neira Ayuso]

Hi,

The following patchset contains Netfilter fixes for net:

1) Add commit and abort set operation to pipapo set abort path.

2) Bail out immediately in case of ENOMEM in nfnetlink batch.

3) Incorrect error path handling when creating a new rule leads to
   dangling pointer in set transaction list.

Please, pull these changes from:

  git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf.git nf-23-06-08

Thanks.

----------------------------------------------------------------

The following changes since commit ab39b113e74751958aac1b125a14ee42bd7d3efd:

  Merge tag 'for-net-2023-06-05' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth (2023-06-06 21:36:57 -0700)

are available in the Git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf.git tags/nf-23-06-08

for you to fetch changes up to 1240eb93f0616b21c675416516ff3d74798fdc97:

  netfilter: nf_tables: incorrect error path handling with NFT_MSG_NEWRULE (2023-06-08 21:49:26 +0200)

----------------------------------------------------------------
netfilter pull request 23-06-08

----------------------------------------------------------------
Pablo Neira Ayuso (3):
      netfilter: nf_tables: integrate pipapo into commit protocol
      netfilter: nfnetlink: skip error delivery on batch in case of ENOMEM
      netfilter: nf_tables: incorrect error path handling with NFT_MSG_NEWRULE

 include/net/netfilter/nf_tables.h |  4 ++-
 net/netfilter/nf_tables_api.c     | 59 ++++++++++++++++++++++++++++++++++++++-
 net/netfilter/nfnetlink.c         |  3 +-
 net/netfilter/nft_set_pipapo.c    | 55 ++++++++++++++++++++++++++----------
 4 files changed, 103 insertions(+), 18 deletions(-)

[PATCH net 1/3] netfilter: nf_tables: integrate pipapo into commit protocol

[Pablo Neira Ayuso]

The pipapo set backend follows copy-on-update approach, maintaining one
clone of the existing datastructure that is being updated. The clone
and current datastructures are swapped via rcu from the commit step.

The existing integration with the commit protocol is flawed because
there is no operation to clean up the clone if the transaction is
aborted. Moreover, the datastructure swap happens on set element
activation.

This patch adds two new operations for sets: commit and abort, these new
operations are invoked from the commit and abort steps, after the
transactions have been digested, and it updates the pipapo set backend
to use it.

This patch adds a new ->pending_update field to sets to maintain a list
of sets that require this new commit and abort operations.

Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
 include/net/netfilter/nf_tables.h |  4 ++-
 net/netfilter/nf_tables_api.c     | 56 +++++++++++++++++++++++++++++++
 net/netfilter/nft_set_pipapo.c    | 55 +++++++++++++++++++++---------
 3 files changed, 99 insertions(+), 16 deletions(-)

diff --git a/include/net/netfilter/nf_tables.h b/include/net/netfilter/nf_tables.h
index 2e24ea1d744c..83db182decc8 100644
--- a/include/net/netfilter/nf_tables.h
+++ b/include/net/netfilter/nf_tables.h
@@ -462,7 +462,8 @@ struct nft_set_ops {
 					       const struct nft_set *set,
 					       const struct nft_set_elem *elem,
 					       unsigned int flags);
-
+	void				(*commit)(const struct nft_set *set);
+	void				(*abort)(const struct nft_set *set);
 	u64				(*privsize)(const struct nlattr * const nla[],
 						    const struct nft_set_desc *desc);
 	bool				(*estimate)(const struct nft_set_desc *desc,
@@ -557,6 +558,7 @@ struct nft_set {
 	u16				policy;
 	u16				udlen;
 	unsigned char			*udata;
+	struct list_head		pending_update;
 	/* runtime data below here */
 	const struct nft_set_ops	*ops ____cacheline_aligned;
 	u16				flags:14,
diff --git a/net/netfilter/nf_tables_api.c b/net/netfilter/nf_tables_api.c
index 0519d45ede6b..3bb0800b3849 100644
--- a/net/netfilter/nf_tables_api.c
+++ b/net/netfilter/nf_tables_api.c
@@ -4919,6 +4919,7 @@ static int nf_tables_newset(struct sk_buff *skb, const struct nfnl_info *info,
 
 	set->num_exprs = num_exprs;
 	set->handle = nf_tables_alloc_handle(table);
+	INIT_LIST_HEAD(&set->pending_update);
 
 	err = nft_trans_set_add(&ctx, NFT_MSG_NEWSET, set);
 	if (err < 0)
@@ -9275,10 +9276,25 @@ static void nf_tables_commit_audit_log(struct list_head *adl, u32 generation)
 	}
 }
 
+static void nft_set_commit_update(struct list_head *set_update_list)
+{
+	struct nft_set *set, *next;
+
+	list_for_each_entry_safe(set, next, set_update_list, pending_update) {
+		list_del_init(&set->pending_update);
+
+		if (!set->ops->commit)
+			continue;
+
+		set->ops->commit(set);
+	}
+}
+
 static int nf_tables_commit(struct net *net, struct sk_buff *skb)
 {
 	struct nftables_pernet *nft_net = nft_pernet(net);
 	struct nft_trans *trans, *next;
+	LIST_HEAD(set_update_list);
 	struct nft_trans_elem *te;
 	struct nft_chain *chain;
 	struct nft_table *table;
@@ -9453,6 +9469,11 @@ static int nf_tables_commit(struct net *net, struct sk_buff *skb)
 			nf_tables_setelem_notify(&trans->ctx, te->set,
 						 &te->elem,
 						 NFT_MSG_NEWSETELEM);
+			if (te->set->ops->commit &&
+			    list_empty(&te->set->pending_update)) {
+				list_add_tail(&te->set->pending_update,
+					      &set_update_list);
+			}
 			nft_trans_destroy(trans);
 			break;
 		case NFT_MSG_DELSETELEM:
@@ -9467,6 +9488,11 @@ static int nf_tables_commit(struct net *net, struct sk_buff *skb)
 				atomic_dec(&te->set->nelems);
 				te->set->ndeact--;
 			}
+			if (te->set->ops->commit &&
+			    list_empty(&te->set->pending_update)) {
+				list_add_tail(&te->set->pending_update,
+					      &set_update_list);
+			}
 			break;
 		case NFT_MSG_NEWOBJ:
 			if (nft_trans_obj_update(trans)) {
@@ -9529,6 +9555,8 @@ static int nf_tables_commit(struct net *net, struct sk_buff *skb)
 		}
 	}
 
+	nft_set_commit_update(&set_update_list);
+
 	nft_commit_notify(net, NETLINK_CB(skb).portid);
 	nf_tables_gen_notify(net, skb, NFT_MSG_NEWGEN);
 	nf_tables_commit_audit_log(&adl, nft_net->base_seq);
@@ -9588,10 +9616,25 @@ static void nf_tables_abort_release(struct nft_trans *trans)
 	kfree(trans);
 }
 
+static void nft_set_abort_update(struct list_head *set_update_list)
+{
+	struct nft_set *set, *next;
+
+	list_for_each_entry_safe(set, next, set_update_list, pending_update) {
+		list_del_init(&set->pending_update);
+
+		if (!set->ops->abort)
+			continue;
+
+		set->ops->abort(set);
+	}
+}
+
 static int __nf_tables_abort(struct net *net, enum nfnl_abort_action action)
 {
 	struct nftables_pernet *nft_net = nft_pernet(net);
 	struct nft_trans *trans, *next;
+	LIST_HEAD(set_update_list);
 	struct nft_trans_elem *te;
 
 	if (action == NFNL_ABORT_VALIDATE &&
@@ -9701,6 +9744,12 @@ static int __nf_tables_abort(struct net *net, enum nfnl_abort_action action)
 			nft_setelem_remove(net, te->set, &te->elem);
 			if (!nft_setelem_is_catchall(te->set, &te->elem))
 				atomic_dec(&te->set->nelems);
+
+			if (te->set->ops->abort &&
+			    list_empty(&te->set->pending_update)) {
+				list_add_tail(&te->set->pending_update,
+					      &set_update_list);
+			}
 			break;
 		case NFT_MSG_DELSETELEM:
 		case NFT_MSG_DESTROYSETELEM:
@@ -9711,6 +9760,11 @@ static int __nf_tables_abort(struct net *net, enum nfnl_abort_action action)
 			if (!nft_setelem_is_catchall(te->set, &te->elem))
 				te->set->ndeact--;
 
+			if (te->set->ops->abort &&
+			    list_empty(&te->set->pending_update)) {
+				list_add_tail(&te->set->pending_update,
+					      &set_update_list);
+			}
 			nft_trans_destroy(trans);
 			break;
 		case NFT_MSG_NEWOBJ:
@@ -9753,6 +9807,8 @@ static int __nf_tables_abort(struct net *net, enum nfnl_abort_action action)
 		}
 	}
 
+	nft_set_abort_update(&set_update_list);
+
 	synchronize_rcu();
 
 	list_for_each_entry_safe_reverse(trans, next,
diff --git a/net/netfilter/nft_set_pipapo.c b/net/netfilter/nft_set_pipapo.c
index 06d46d182634..15e451dc3fc4 100644
--- a/net/netfilter/nft_set_pipapo.c
+++ b/net/netfilter/nft_set_pipapo.c
@@ -1600,17 +1600,10 @@ static void pipapo_free_fields(struct nft_pipapo_match *m)
 	}
 }
 
-/**
- * pipapo_reclaim_match - RCU callback to free fields from old matching data
- * @rcu:	RCU head
- */
-static void pipapo_reclaim_match(struct rcu_head *rcu)
+static void pipapo_free_match(struct nft_pipapo_match *m)
 {
-	struct nft_pipapo_match *m;
 	int i;
 
-	m = container_of(rcu, struct nft_pipapo_match, rcu);
-
 	for_each_possible_cpu(i)
 		kfree(*per_cpu_ptr(m->scratch, i));
 
@@ -1625,7 +1618,19 @@ static void pipapo_reclaim_match(struct rcu_head *rcu)
 }
 
 /**
- * pipapo_commit() - Replace lookup data with current working copy
+ * pipapo_reclaim_match - RCU callback to free fields from old matching data
+ * @rcu:	RCU head
+ */
+static void pipapo_reclaim_match(struct rcu_head *rcu)
+{
+	struct nft_pipapo_match *m;
+
+	m = container_of(rcu, struct nft_pipapo_match, rcu);
+	pipapo_free_match(m);
+}
+
+/**
+ * nft_pipapo_commit() - Replace lookup data with current working copy
  * @set:	nftables API set representation
  *
  * While at it, check if we should perform garbage collection on the working
@@ -1635,7 +1640,7 @@ static void pipapo_reclaim_match(struct rcu_head *rcu)
  * We also need to create a new working copy for subsequent insertions and
  * deletions.
  */
-static void pipapo_commit(const struct nft_set *set)
+static void nft_pipapo_commit(const struct nft_set *set)
 {
 	struct nft_pipapo *priv = nft_set_priv(set);
 	struct nft_pipapo_match *new_clone, *old;
@@ -1660,6 +1665,26 @@ static void pipapo_commit(const struct nft_set *set)
 	priv->clone = new_clone;
 }
 
+static void nft_pipapo_abort(const struct nft_set *set)
+{
+	struct nft_pipapo *priv = nft_set_priv(set);
+	struct nft_pipapo_match *new_clone, *m;
+
+	if (!priv->dirty)
+		return;
+
+	m = rcu_dereference(priv->match);
+
+	new_clone = pipapo_clone(m);
+	if (IS_ERR(new_clone))
+		return;
+
+	priv->dirty = false;
+
+	pipapo_free_match(priv->clone);
+	priv->clone = new_clone;
+}
+
 /**
  * nft_pipapo_activate() - Mark element reference as active given key, commit
  * @net:	Network namespace
@@ -1667,8 +1692,7 @@ static void pipapo_commit(const struct nft_set *set)
  * @elem:	nftables API element representation containing key data
  *
  * On insertion, elements are added to a copy of the matching data currently
- * in use for lookups, and not directly inserted into current lookup data, so
- * we'll take care of that by calling pipapo_commit() here. Both
+ * in use for lookups, and not directly inserted into current lookup data. Both
  * nft_pipapo_insert() and nft_pipapo_activate() are called once for each
  * element, hence we can't purpose either one as a real commit operation.
  */
@@ -1684,8 +1708,6 @@ static void nft_pipapo_activate(const struct net *net,
 
 	nft_set_elem_change_active(net, set, &e->ext);
 	nft_set_elem_clear_busy(&e->ext);
-
-	pipapo_commit(set);
 }
 
 /**
@@ -1931,7 +1953,6 @@ static void nft_pipapo_remove(const struct net *net, const struct nft_set *set,
 		if (i == m->field_count) {
 			priv->dirty = true;
 			pipapo_drop(m, rulemap);
-			pipapo_commit(set);
 			return;
 		}
 
@@ -2230,6 +2251,8 @@ const struct nft_set_type nft_set_pipapo_type = {
 		.init		= nft_pipapo_init,
 		.destroy	= nft_pipapo_destroy,
 		.gc_init	= nft_pipapo_gc_init,
+		.commit		= nft_pipapo_commit,
+		.abort		= nft_pipapo_abort,
 		.elemsize	= offsetof(struct nft_pipapo_elem, ext),
 	},
 };
@@ -2252,6 +2275,8 @@ const struct nft_set_type nft_set_pipapo_avx2_type = {
 		.init		= nft_pipapo_init,
 		.destroy	= nft_pipapo_destroy,
 		.gc_init	= nft_pipapo_gc_init,
+		.commit		= nft_pipapo_commit,
+		.abort		= nft_pipapo_abort,
 		.elemsize	= offsetof(struct nft_pipapo_elem, ext),
 	},
 };
-- 
2.30.2

Re: [PATCH net 1/3] netfilter: nf_tables: integrate pipapo into commit protocol

[Simon Horman]

On Thu, Jun 08, 2023 at 09:57:04PM +0200, Pablo Neira Ayuso wrote:
> The pipapo set backend follows copy-on-update approach, maintaining one
> clone of the existing datastructure that is being updated. The clone
> and current datastructures are swapped via rcu from the commit step.
> 
> The existing integration with the commit protocol is flawed because
> there is no operation to clean up the clone if the transaction is
> aborted. Moreover, the datastructure swap happens on set element
> activation.
> 
> This patch adds two new operations for sets: commit and abort, these new
> operations are invoked from the commit and abort steps, after the
> transactions have been digested, and it updates the pipapo set backend
> to use it.
> 
> This patch adds a new ->pending_update field to sets to maintain a list
> of sets that require this new commit and abort operations.
> 
> Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges")
> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Hi Pablo,

some suggestions of some trivial follow-up items from my side.
No need for these to hold up progress of the patchset.

> ---
>  include/net/netfilter/nf_tables.h |  4 ++-
>  net/netfilter/nf_tables_api.c     | 56 +++++++++++++++++++++++++++++++
>  net/netfilter/nft_set_pipapo.c    | 55 +++++++++++++++++++++---------
>  3 files changed, 99 insertions(+), 16 deletions(-)
> 
> diff --git a/include/net/netfilter/nf_tables.h b/include/net/netfilter/nf_tables.h
> index 2e24ea1d744c..83db182decc8 100644
> --- a/include/net/netfilter/nf_tables.h
> +++ b/include/net/netfilter/nf_tables.h
> @@ -462,7 +462,8 @@ struct nft_set_ops {
>  					       const struct nft_set *set,
>  					       const struct nft_set_elem *elem,
>  					       unsigned int flags);
> -
> +	void				(*commit)(const struct nft_set *set);
> +	void				(*abort)(const struct nft_set *set);

As a follow-up, these could be added to the kdoc for nft_set_ops.

>  	u64				(*privsize)(const struct nlattr * const nla[],
>  						    const struct nft_set_desc *desc);
>  	bool				(*estimate)(const struct nft_set_desc *desc,
> @@ -557,6 +558,7 @@ struct nft_set {
>  	u16				policy;
>  	u16				udlen;
>  	unsigned char			*udata;
> +	struct list_head		pending_update;

Likewise, as a follow-up, pending_update could be added to the kdoc for
nft_set.

>  	/* runtime data below here */
>  	const struct nft_set_ops	*ops ____cacheline_aligned;
>  	u16				flags:14,

...

Re: [PATCH net 1/3] netfilter: nf_tables: integrate pipapo into commit protocol

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net.git (main)
by Pablo Neira Ayuso <pablo@netfilter.org>:

On Thu,  8 Jun 2023 21:57:04 +0200 you wrote:
> The pipapo set backend follows copy-on-update approach, maintaining one
> clone of the existing datastructure that is being updated. The clone
> and current datastructures are swapped via rcu from the commit step.
> 
> The existing integration with the commit protocol is flawed because
> there is no operation to clean up the clone if the transaction is
> aborted. Moreover, the datastructure swap happens on set element
> activation.
> 
> [...]

Here is the summary with links:
  - [net,1/3] netfilter: nf_tables: integrate pipapo into commit protocol
    https://git.kernel.org/netdev/net/c/212ed75dc5fb
  - [net,2/3] netfilter: nfnetlink: skip error delivery on batch in case of ENOMEM
    https://git.kernel.org/netdev/net/c/a1a64a151dae
  - [net,3/3] netfilter: nf_tables: incorrect error path handling with NFT_MSG_NEWRULE
    https://git.kernel.org/netdev/net/c/1240eb93f061

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

[PATCH net 2/3] netfilter: nfnetlink: skip error delivery on batch in case of ENOMEM

[Pablo Neira Ayuso]

If caller reports ENOMEM, then stop iterating over the batch and send a
single netlink message to userspace to report OOM.

Fixes: cbb8125eb40b ("netfilter: nfnetlink: deliver netlink errors on batch completion")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
 net/netfilter/nfnetlink.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/netfilter/nfnetlink.c b/net/netfilter/nfnetlink.c
index ae7146475d17..c9fbe0f707b5 100644
--- a/net/netfilter/nfnetlink.c
+++ b/net/netfilter/nfnetlink.c
@@ -533,7 +533,8 @@ static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
 			 * processed, this avoids that the same error is
 			 * reported several times when replaying the batch.
 			 */
-			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
+			if (err == -ENOMEM ||
+			    nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
 				/* We failed to enqueue an error, reset the
 				 * list of errors and send OOM to userspace
 				 * pointing to the batch header.
-- 
2.30.2

[PATCH net 3/3] netfilter: nf_tables: incorrect error path handling with NFT_MSG_NEWRULE

[Pablo Neira Ayuso]

In case of error when adding a new rule that refers to an anonymous set,
deactivate expressions via NFT_TRANS_PREPARE state, not NFT_TRANS_RELEASE.
Thus, the lookup expression marks anonymous sets as inactive in the next
generation to ensure it is not reachable in this transaction anymore and
decrement the set refcount as introduced by c1592a89942e ("netfilter:
nf_tables: deactivate anonymous set from preparation phase"). The abort
step takes care of undoing the anonymous set.

This is also consistent with rule deletion, where NFT_TRANS_PREPARE is
used. Note that this error path is exercised in the preparation step of
the commit protocol. This patch replaces nf_tables_rule_release() by the
deactivate and destroy calls, this time with NFT_TRANS_PREPARE.

Due to this incorrect error handling, it is possible to access a
dangling pointer to the anonymous set that remains in the transaction
list.

[1009.379054] BUG: KASAN: use-after-free in nft_set_lookup_global+0x147/0x1a0 [nf_tables]
[1009.379106] Read of size 8 at addr ffff88816c4c8020 by task nft-rule-add/137110
[1009.379116] CPU: 7 PID: 137110 Comm: nft-rule-add Not tainted 6.4.0-rc4+ #256
[1009.379128] Call Trace:
[1009.379132]  <TASK>
[1009.379135]  dump_stack_lvl+0x33/0x50
[1009.379146]  ? nft_set_lookup_global+0x147/0x1a0 [nf_tables]
[1009.379191]  print_address_description.constprop.0+0x27/0x300
[1009.379201]  kasan_report+0x107/0x120
[1009.379210]  ? nft_set_lookup_global+0x147/0x1a0 [nf_tables]
[1009.379255]  nft_set_lookup_global+0x147/0x1a0 [nf_tables]
[1009.379302]  nft_lookup_init+0xa5/0x270 [nf_tables]
[1009.379350]  nf_tables_newrule+0x698/0xe50 [nf_tables]
[1009.379397]  ? nf_tables_rule_release+0xe0/0xe0 [nf_tables]
[1009.379441]  ? kasan_unpoison+0x23/0x50
[1009.379450]  nfnetlink_rcv_batch+0x97c/0xd90 [nfnetlink]
[1009.379470]  ? nfnetlink_rcv_msg+0x480/0x480 [nfnetlink]
[1009.379485]  ? __alloc_skb+0xb8/0x1e0
[1009.379493]  ? __alloc_skb+0xb8/0x1e0
[1009.379502]  ? entry_SYSCALL_64_after_hwframe+0x46/0xb0
[1009.379509]  ? unwind_get_return_address+0x2a/0x40
[1009.379517]  ? write_profile+0xc0/0xc0
[1009.379524]  ? avc_lookup+0x8f/0xc0
[1009.379532]  ? __rcu_read_unlock+0x43/0x60

Fixes: 958bee14d071 ("netfilter: nf_tables: use new transaction infrastructure to handle sets")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
 net/netfilter/nf_tables_api.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/netfilter/nf_tables_api.c b/net/netfilter/nf_tables_api.c
index 3bb0800b3849..69bceefaa5c8 100644
--- a/net/netfilter/nf_tables_api.c
+++ b/net/netfilter/nf_tables_api.c
@@ -3844,7 +3844,8 @@ static int nf_tables_newrule(struct sk_buff *skb, const struct nfnl_info *info,
 	if (flow)
 		nft_flow_rule_destroy(flow);
 err_release_rule:
-	nf_tables_rule_release(&ctx, rule);
+	nft_rule_expr_deactivate(&ctx, rule, NFT_TRANS_PREPARE);
+	nf_tables_rule_destroy(&ctx, rule);
 err_release_expr:
 	for (i = 0; i < n; i++) {
 		if (expr_info[i].ops) {
-- 
2.30.2