[PATCH nf] nf_tables: nft_dynset: fix possible stateful expression memleak in error path

[PATCH nf] nf_tables: nft_dynset: fix possible stateful expression memleak in error path

[Pablo Neira Ayuso]

If cloning the second stateful expression in the element via GFP_ATOMIC
fails, then the first stateful expression remains in place without being
released.

   unreferenced object (percpu) 0x607b97e9cab8 (size 16):
     comm "softirq", pid 0, jiffies 4294931867
     hex dump (first 16 bytes on cpu 3):
       00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
     backtrace (crc 0):
       pcpu_alloc_noprof+0x453/0xd80
       nft_counter_clone+0x9c/0x190 [nf_tables]
       nft_expr_clone+0x8f/0x1b0 [nf_tables]
       nft_dynset_new+0x2cb/0x5f0 [nf_tables]
       nft_rhash_update+0x236/0x11c0 [nf_tables]
       nft_dynset_eval+0x11f/0x670 [nf_tables]
       nft_do_chain+0x253/0x1700 [nf_tables]
       nft_do_chain_ipv4+0x18d/0x270 [nf_tables]
       nf_hook_slow+0xaa/0x1e0
       ip_local_deliver+0x209/0x330

Pass NULL to nft_set_elem_expr_destroy() given stateful expressions do
not require context at this stage.

Fixes: 563125a73ac3 ("netfilter: nftables: generalize set extension to support for several expressions")
Reported-by: Gurpreet Shergill <giki.shergill@proton.me>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
---
Compile-tested only at this stage.

 include/net/netfilter/nf_tables.h | 2 ++
 net/netfilter/nf_tables_api.c     | 4 ++--
 net/netfilter/nft_dynset.c        | 7 ++++++-
 3 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/include/net/netfilter/nf_tables.h b/include/net/netfilter/nf_tables.h
index ea6f29ad7888..3c8a60ec1cc4 100644
--- a/include/net/netfilter/nf_tables.h
+++ b/include/net/netfilter/nf_tables.h
@@ -878,6 +878,8 @@ struct nft_elem_priv *nft_set_elem_init(const struct nft_set *set,
 					u64 timeout, u64 expiration, gfp_t gfp);
 int nft_set_elem_expr_clone(const struct nft_ctx *ctx, struct nft_set *set,
 			    struct nft_expr *expr_array[]);
+void nft_set_elem_expr_destroy(const struct nft_ctx *ctx,
+			       struct nft_set_elem_expr *elem_expr);
 void nft_set_elem_destroy(const struct nft_set *set,
 			  const struct nft_elem_priv *elem_priv,
 			  bool destroy_expr);
diff --git a/net/netfilter/nf_tables_api.c b/net/netfilter/nf_tables_api.c
index 15801a9a099e..2f19c155069e 100644
--- a/net/netfilter/nf_tables_api.c
+++ b/net/netfilter/nf_tables_api.c
@@ -6747,8 +6747,8 @@ static void __nft_set_elem_expr_destroy(const struct nft_ctx *ctx,
 	}
 }
 
-static void nft_set_elem_expr_destroy(const struct nft_ctx *ctx,
-				      struct nft_set_elem_expr *elem_expr)
+void nft_set_elem_expr_destroy(const struct nft_ctx *ctx,
+			       struct nft_set_elem_expr *elem_expr)
 {
 	struct nft_expr *expr;
 	u32 size;
diff --git a/net/netfilter/nft_dynset.c b/net/netfilter/nft_dynset.c
index 7807d8129664..31c5a5b52ce1 100644
--- a/net/netfilter/nft_dynset.c
+++ b/net/netfilter/nft_dynset.c
@@ -36,12 +36,17 @@ static int nft_dynset_expr_setup(const struct nft_dynset *priv,
 	for (i = 0; i < priv->num_exprs; i++) {
 		expr = nft_setelem_expr_at(elem_expr, elem_expr->size);
 		if (nft_expr_clone(expr, priv->expr_array[i], GFP_ATOMIC) < 0)
-			return -1;
+			goto err_out;
 
 		elem_expr->size += priv->expr_array[i]->ops->size;
 	}
 
 	return 0;
+err_out:
+	/* Stateful expression do not need context, pass NULL. */
+	nft_set_elem_expr_destroy(NULL, elem_expr);
+
+	return -1;
 }
 
 struct nft_elem_priv *nft_dynset_new(struct nft_set *set,
-- 
2.47.3

Re: [PATCH nf] nf_tables: nft_dynset: fix possible stateful expression memleak in error path

[Florian Westphal]

Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> If cloning the second stateful expression in the element via GFP_ATOMIC
> fails, then the first stateful expression remains in place without being
> released.
> 
>    unreferenced object (percpu) 0x607b97e9cab8 (size 16):
>      comm "softirq", pid 0, jiffies 4294931867
>      hex dump (first 16 bytes on cpu 3):
>        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>      backtrace (crc 0):
>        pcpu_alloc_noprof+0x453/0xd80
>        nft_counter_clone+0x9c/0x190 [nf_tables]
>        nft_expr_clone+0x8f/0x1b0 [nf_tables]
>        nft_dynset_new+0x2cb/0x5f0 [nf_tables]
>        nft_rhash_update+0x236/0x11c0 [nf_tables]
>        nft_dynset_eval+0x11f/0x670 [nf_tables]
>        nft_do_chain+0x253/0x1700 [nf_tables]
>        nft_do_chain_ipv4+0x18d/0x270 [nf_tables]
>        nf_hook_slow+0xaa/0x1e0
>        ip_local_deliver+0x209/0x330
> 
> Pass NULL to nft_set_elem_expr_destroy() given stateful expressions do
> not require context at this stage.

static void nft_connlimit_do_destroy(const struct nft_ctx *ctx,
                                     struct nft_connlimit *priv)
{
        nf_ct_netns_put(ctx->net, ctx->family);
        nf_conncount_cache_free(priv->list);
        kfree(priv->list);
}

I think minimal fake context could work though, the clone wasn't
exposed to other cpus yet.

Other than this patch looks correct to me.

Re: [PATCH nf] nf_tables: nft_dynset: fix possible stateful expression memleak in error path

[Pablo Neira Ayuso]

On Thu, Mar 12, 2026 at 12:02:37PM +0100, Florian Westphal wrote:
> Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > If cloning the second stateful expression in the element via GFP_ATOMIC
> > fails, then the first stateful expression remains in place without being
> > released.
> > 
> >    unreferenced object (percpu) 0x607b97e9cab8 (size 16):
> >      comm "softirq", pid 0, jiffies 4294931867
> >      hex dump (first 16 bytes on cpu 3):
> >        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> >      backtrace (crc 0):
> >        pcpu_alloc_noprof+0x453/0xd80
> >        nft_counter_clone+0x9c/0x190 [nf_tables]
> >        nft_expr_clone+0x8f/0x1b0 [nf_tables]
> >        nft_dynset_new+0x2cb/0x5f0 [nf_tables]
> >        nft_rhash_update+0x236/0x11c0 [nf_tables]
> >        nft_dynset_eval+0x11f/0x670 [nf_tables]
> >        nft_do_chain+0x253/0x1700 [nf_tables]
> >        nft_do_chain_ipv4+0x18d/0x270 [nf_tables]
> >        nf_hook_slow+0xaa/0x1e0
> >        ip_local_deliver+0x209/0x330
> > 
> > Pass NULL to nft_set_elem_expr_destroy() given stateful expressions do
> > not require context at this stage.
> 
> static void nft_connlimit_do_destroy(const struct nft_ctx *ctx,
>                                      struct nft_connlimit *priv)
> {
>         nf_ct_netns_put(ctx->net, ctx->family);
>         nf_conncount_cache_free(priv->list);
>         kfree(priv->list);
> }
> 
> I think minimal fake context could work though, the clone wasn't
> exposed to other cpus yet.
> 
> Other than this patch looks correct to me.

It is following this path:

static void nft_connlimit_destroy_clone(const struct nft_ctx *ctx,
                                        const struct nft_expr *expr)
{
        struct nft_connlimit *priv = nft_expr_priv(expr);
 
        nf_conncount_cache_free(priv->list);
        kfree(priv->list);
}

But I will post a v2 setting nft_ctx anyway.

Re: [PATCH nf] nf_tables: nft_dynset: fix possible stateful expression memleak in error path

[Florian Westphal]

Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> On Thu, Mar 12, 2026 at 12:02:37PM +0100, Florian Westphal wrote:
> > Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > > If cloning the second stateful expression in the element via GFP_ATOMIC
> > > fails, then the first stateful expression remains in place without being
> > > released.
> > > 
> > >    unreferenced object (percpu) 0x607b97e9cab8 (size 16):
> > >      comm "softirq", pid 0, jiffies 4294931867
> > >      hex dump (first 16 bytes on cpu 3):
> > >        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> > >      backtrace (crc 0):
> > >        pcpu_alloc_noprof+0x453/0xd80
> > >        nft_counter_clone+0x9c/0x190 [nf_tables]
> > >        nft_expr_clone+0x8f/0x1b0 [nf_tables]
> > >        nft_dynset_new+0x2cb/0x5f0 [nf_tables]
> > >        nft_rhash_update+0x236/0x11c0 [nf_tables]
> > >        nft_dynset_eval+0x11f/0x670 [nf_tables]
> > >        nft_do_chain+0x253/0x1700 [nf_tables]
> > >        nft_do_chain_ipv4+0x18d/0x270 [nf_tables]
> > >        nf_hook_slow+0xaa/0x1e0
> > >        ip_local_deliver+0x209/0x330
> > > 
> > > Pass NULL to nft_set_elem_expr_destroy() given stateful expressions do
> > > not require context at this stage.
> > 
> > static void nft_connlimit_do_destroy(const struct nft_ctx *ctx,
> >                                      struct nft_connlimit *priv)
> > {
> >         nf_ct_netns_put(ctx->net, ctx->family);
> >         nf_conncount_cache_free(priv->list);
> >         kfree(priv->list);
> > }
> > 
> > I think minimal fake context could work though, the clone wasn't
> > exposed to other cpus yet.
> > 
> > Other than this patch looks correct to me.
> 
> It is following this path:
>
> static void nft_connlimit_destroy_clone(const struct nft_ctx *ctx,
>                                         const struct nft_expr *expr)
> {
>         struct nft_connlimit *priv = nft_expr_priv(expr);
>  
>         nf_conncount_cache_free(priv->list);
>         kfree(priv->list);
> }

Oh, I see -- you are right, this follows the destroy_expr special case.

> But I will post a v2 setting nft_ctx anyway.

Thanks, I find it hard to assert correctness otherwise.