Re: [regression] TCP_MD5SIG on established sockets

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- On Jun 30, 2020, at 11:36 PM, Eric Dumazet edumazet@google.com wrote:

> On Tue, Jun 30, 2020 at 7:59 PM Herbert Xu <herbert@gondor.apana.org.au> wrote:
>>
>> On Tue, Jun 30, 2020 at 07:30:43PM -0700, Eric Dumazet wrote:
>> >
>> > I made this clear in the changelog, do we want comments all over the places ?
>> > Do not get me wrong, we had this bug for years and suddenly this is a
>> > big deal...
>>
>> I thought you were adding a new pair of smp_rmb/smp_wmb.  If they
>> already exist in the code then I agree it's not a big deal.  But
>> adding a new pair of bogus smp_Xmb's is bad for maintenance.
>>
> 
> If I knew so many people were excited about TCP / MD5, I would have
> posted all my patches on lkml ;)
> 
> Without the smp_wmb() we would still need something to prevent KMSAN
> from detecting that we read uninitialized bytes,
> if key->keylen is increased.  (initial content of key->key[] is garbage)
> 
> Something like this :

The approach below looks good to me, but you'll also need to annotate
both tcp_md5_hash_key and tcp_md5_do_add with __no_kcsan or use
data_race(expr) to let the concurrency sanitizer know that there is
a known data race which is there on purpose (triggered by memcpy in tcp_md5_do_add
and somewhere within crypto_ahash_update). See Documentation/dev-tools/kcsan.rst
for details.

Thanks,

Mathieu

> 
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index
> f111660453241692a17c881dd6dc2910a1236263..c3af8180c7049d5c4987bf5c67e4aff2ed6967c9
> 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -4033,11 +4033,9 @@ EXPORT_SYMBOL(tcp_md5_hash_skb_data);
> 
> int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct
> tcp_md5sig_key *key)
> {
> -       u8 keylen = key->keylen;
> +       u8 keylen = READ_ONCE(key->keylen); /* paired with
> WRITE_ONCE() in tcp_md5_do_add */
>        struct scatterlist sg;
> 
> -       smp_rmb(); /* paired with smp_wmb() in tcp_md5_do_add() */
> -
>        sg_init_one(&sg, key->key, keylen);
>        ahash_request_set_crypt(hp->md5_req, &sg, NULL, keylen);
>        return crypto_ahash_update(hp->md5_req);
> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> index
> 99916fcc15ca0be12c2c133ff40516f79e6fdf7f..0d08e0134335a21d23702e6a5c24a0f2b3c61c6f
> 100644
> --- a/net/ipv4/tcp_ipv4.c
> +++ b/net/ipv4/tcp_ipv4.c
> @@ -1114,9 +1114,13 @@ int tcp_md5_do_add(struct sock *sk, const union
> tcp_md5_addr *addr,
>                /* Pre-existing entry - just update that one. */
>                memcpy(key->key, newkey, newkeylen);
> 
> -               smp_wmb(); /* pairs with smp_rmb() in tcp_md5_hash_key() */
> +               /* Pairs with READ_ONCE() in tcp_md5_hash_key().
> +                * Also note that a reader could catch new key->keylen value
> +                * but old key->key[], this is the reason we use __GFP_ZERO
> +                * at sock_kmalloc() time below these lines.
> +                */
> +               WRITE_ONCE(key->keylen, newkeylen);
> 
> -               key->keylen = newkeylen;
>                return 0;
>        }
> 
> @@ -1132,7 +1136,7 @@ int tcp_md5_do_add(struct sock *sk, const union
> tcp_md5_addr *addr,
>                rcu_assign_pointer(tp->md5sig_info, md5sig);
>        }
> 
> -       key = sock_kmalloc(sk, sizeof(*key), gfp);
> +       key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
>        if (!key)
>                return -ENOMEM;
>         if (!tcp_alloc_md5sig_pool()) {

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com