Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.37 as permitted sender) client-ip=23.128.96.37;
Message-ID: <20aed7c604115c7f346cdbbe848c37feded3cf21.camel@redhat.com>
Subject: Re: [PATCH] net/tls: Fix slab-use-after-free in tls_encrypt_done
From:   Paolo Abeni <pabeni@redhat.com>
To:     Juntong Deng <juntong.deng@outlook.com>, borisp@nvidia.com,
        john.fastabend@gmail.com, kuba@kernel.org, davem@davemloft.net,
        edumazet@google.com
Cc:     netdev@vger.kernel.org,
        linux-kernel-mentees@lists.linuxfoundation.org,
        linux-kernel@vger.kernel.org,
        syzbot+29c22ea2d6b2c5fd2eae@syzkaller.appspotmail.com
Date:   Tue, 17 Oct 2023 12:31:36 +0200
In-Reply-To: <VI1P193MB0752428D259D066379242BD099D3A@VI1P193MB0752.EURP193.PROD.OUTLOOK.COM>
References: <VI1P193MB0752428D259D066379242BD099D3A@VI1P193MB0752.EURP193.PROD.OUTLOOK.COM>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Thu, 2023-10-12 at 19:02 +0800, Juntong Deng wrote:
> In the current implementation, ctx->async_wait.completion is completed
> after spin_lock_bh, which causes tls_sw_release_resources_tx to
> continue executing and return to tls_sk_proto_cleanup, then return
> to tls_sk_proto_close, and after that enter tls_sw_free_ctx_tx to kfree
> the entire struct tls_context (including ctx->encrypt_compl_lock).
>=20
> Since ctx->encrypt_compl_lock has been freed, subsequent spin_unlock_bh
> will result in slab-use-after-free error. Due to SMP, even using
> spin_lock_bh does not prevent tls_sw_release_resources_tx from continuing
> on other CPUs. After tls_sw_release_resources_tx is woken up, there is no
> attempt to hold ctx->encrypt_compl_lock again, therefore everything
> described above is possible.
>=20
> The fix is to put complete(&ctx->async_wait.completion) after
> spin_unlock_bh, making the release after the unlock. Since complete is
> only executed if pending is 0, which means this is the last record, there
> is no need to worry about race condition causing duplicate completes.
>=20
> Reported-by: syzbot+29c22ea2d6b2c5fd2eae@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=3D29c22ea2d6b2c5fd2eae
> Signed-off-by: Juntong Deng <juntong.deng@outlook.com>

Have you tested this patch vs the syzbot reproducer?

I think the following race is still present:

CPU0                            CPU1
tls_sw_release_resources_tx     tls_encrypt_done
spin_lock_bh
spin_unlock_bh
                                spin_lock_bh
                                spin_unlock_bh
                                complete

wait
// ...
tls_sk_proto_close

                                test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx=
_bitmask
				// UaF

regardless of 'complete()' being invoked before or after the
'spin_unlock_bh()'.

Paolo