[PATCH] netfilter: mark racy access on ext->gen_id

[PATCH] netfilter: mark racy access on ext->gen_id

[linke li]

In __nf_ct_ext_find(), ext->gen_id can be changed by 
nf_ct_ext_valid_post(), using WRITE_ONCE. Mark data races on ext->gen_id
as benign using READ_ONCE. 

This patch is aimed at reducing the number of benign races reported by
KCSAN in order to focus future debugging effort on harmful races.

Signed-off-by: linke li <lilinke99@qq.com>
---
 net/netfilter/nf_conntrack_extend.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/netfilter/nf_conntrack_extend.c b/net/netfilter/nf_conntrack_extend.c
index dd62cc12e775..7f1a5e5f6646 100644
--- a/net/netfilter/nf_conntrack_extend.c
+++ b/net/netfilter/nf_conntrack_extend.c
@@ -141,7 +141,7 @@ void *__nf_ct_ext_find(const struct nf_ct_ext *ext, u8 id)
 	if (!__nf_ct_ext_exist(ext, id))
 		return NULL;
 
-	if (this_id == 0 || ext->gen_id == gen_id)
+	if (this_id == 0 || READ_ONCE(ext->gen_id) == gen_id)
 		return (void *)ext + ext->offset[id];
 
 	return NULL;
-- 
2.39.3 (Apple Git-146)

Re: [PATCH] netfilter: mark racy access on ext->gen_id

[Florian Westphal]

linke li <lilinke99@qq.com> wrote:
> In __nf_ct_ext_find(), ext->gen_id can be changed by 
> nf_ct_ext_valid_post(), using WRITE_ONCE. Mark data races on ext->gen_id
> as benign using READ_ONCE. 
> 
> This patch is aimed at reducing the number of benign races reported by
> KCSAN in order to focus future debugging effort on harmful races.

Acked-by: Florian Westphal <fw@strlen.de>

Should proably go to nf tree.

Re: [PATCH] netfilter: mark racy access on ext->gen_id

[Pablo Neira Ayuso]

On Tue, Apr 23, 2024 at 02:03:09PM +0200, Florian Westphal wrote:
> linke li <lilinke99@qq.com> wrote:
> > In __nf_ct_ext_find(), ext->gen_id can be changed by 
> > nf_ct_ext_valid_post(), using WRITE_ONCE. Mark data races on ext->gen_id
> > as benign using READ_ONCE. 
> > 
> > This patch is aimed at reducing the number of benign races reported by
> > KCSAN in order to focus future debugging effort on harmful races.
> 
> Acked-by: Florian Westphal <fw@strlen.de>
> 
> Should proably go to nf tree.

Can I get a Fixes: tag for this one?

Re: [PATCH] netfilter: mark racy access on ext->gen_id

[Pablo Neira Ayuso]

On Tue, Apr 23, 2024 at 07:50:22PM +0800, linke li wrote:
> In __nf_ct_ext_find(), ext->gen_id can be changed by 
> nf_ct_ext_valid_post(), using WRITE_ONCE. Mark data races on ext->gen_id
> as benign using READ_ONCE. 
> 
> This patch is aimed at reducing the number of benign races reported by
> KCSAN in order to focus future debugging effort on harmful races.

There are a more uses ext->gen_id in the code, my understanding this
patch is just a stub.

Re: [PATCH] netfilter: mark racy access on ext->gen_id

[Eric Dumazet]

On Thu, Apr 25, 2024 at 11:36 AM Pablo Neira Ayuso <pablo@netfilter.org> wrote:
>
> On Tue, Apr 23, 2024 at 07:50:22PM +0800, linke li wrote:
> > In __nf_ct_ext_find(), ext->gen_id can be changed by
> > nf_ct_ext_valid_post(), using WRITE_ONCE. Mark data races on ext->gen_id
> > as benign using READ_ONCE.
> >
> > This patch is aimed at reducing the number of benign races reported by
> > KCSAN in order to focus future debugging effort on harmful races.
>
> There are a more uses ext->gen_id in the code, my understanding this
> patch is just a stub.

Anyway, ext->gen_id was already read and stored in @this_id

I would probably avoid reading it a second time.

diff --git a/net/netfilter/nf_conntrack_extend.c
b/net/netfilter/nf_conntrack_extend.c
index dd62cc12e7750734fec9be8a90fd0defcbc815e0..747797b20bc7417a2b7270d84f62d24991a4b982
100644
--- a/net/netfilter/nf_conntrack_extend.c
+++ b/net/netfilter/nf_conntrack_extend.c
@@ -141,7 +141,7 @@ void *__nf_ct_ext_find(const struct nf_ct_ext *ext, u8 id)
        if (!__nf_ct_ext_exist(ext, id))
                return NULL;

-       if (this_id == 0 || ext->gen_id == gen_id)
+       if (this_id == 0 || this_id == gen_id)
                return (void *)ext + ext->offset[id];

        return NULL;

Re: [PATCH] netfilter: mark racy access on ext->gen_id

[Florian Westphal]

Eric Dumazet <edumazet@google.com> wrote:
> > There are a more uses ext->gen_id in the code, my understanding this
> > patch is just a stub.
> 
> Anyway, ext->gen_id was already read and stored in @this_id
> 
> I would probably avoid reading it a second time.
 
> diff --git a/net/netfilter/nf_conntrack_extend.c
> b/net/netfilter/nf_conntrack_extend.c
> index dd62cc12e7750734fec9be8a90fd0defcbc815e0..747797b20bc7417a2b7270d84f62d24991a4b982
> 100644
> --- a/net/netfilter/nf_conntrack_extend.c
> +++ b/net/netfilter/nf_conntrack_extend.c
> @@ -141,7 +141,7 @@ void *__nf_ct_ext_find(const struct nf_ct_ext *ext, u8 id)
>         if (!__nf_ct_ext_exist(ext, id))
>                 return NULL;
> 
> -       if (this_id == 0 || ext->gen_id == gen_id)
> +       if (this_id == 0 || this_id == gen_id)
>                 return (void *)ext + ext->offset[id];
> 
>         return NULL;
> 

Right, that should work, unconfirmed entries are not exposed to other
cpus and confirmed entries get their id set to 0.