Re: [PATCH net 5/5] tls: don't decrypt the next record if it's of a different type

[Jakub Kicinski <kuba@kernel.org>]

On Wed,  6 Sep 2023 19:08:35 +0200 Sabrina Dubroca wrote:
> If the next record is of a different type, we won't copy it to
> userspace in this round, tls_record_content_type will stop us just
> after decryption. Skip decryption until the next recvmsg() call.
> 
> This fixes a use-after-free when a data record is decrypted
> asynchronously but doesn't fill the userspace buffer, and the next
> record is non-data, for example in the bad_cmsg selftest.

What's the UAF on?

> diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> index f80a2ea1dd7e..86b835b15872 100644
> --- a/net/tls/tls_sw.c
> +++ b/net/tls/tls_sw.c
> @@ -2010,6 +2010,9 @@ int tls_sw_recvmsg(struct sock *sk,
>  		else
>  			darg.async = false;
>  
> +		if (ctx->async_capable && control && tlm->control != control)
> +			goto recv_end;