netdev - Re: [PATCH V2 net 1/1] net/tls(TLS_SW): Fix list_del double free caused by a race condition in tls_tx

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAOrEdsk6P=HWfK-mKyLt7=tZh342gZrRKwOH9f6ntkNyya-4fA@mail.gmail.com>
Date:   Sat, 21 Sep 2019 23:19:20 -0400
From:   Pooja Trivedi <poojatrivedi@...il.com>
To:     Jakub Kicinski <jakub.kicinski@...ronome.com>
Cc:     netdev@...r.kernel.org, davem@...emloft.net, daniel@...earbox.net,
        john.fastabend@...il.com, davejwatson@...com, aviadye@...lanox.com,
        borisp@...lanox.com, Pooja Trivedi <pooja.trivedi@...ckpath.com>,
        Mallesham Jatharakonda <mallesh537@...il.com>
Subject: Re: [PATCH V2 net 1/1] net/tls(TLS_SW): Fix list_del double free
 caused by a race condition in tls_tx_records

On Wed, Sep 18, 2019 at 5:45 PM Jakub Kicinski
<jakub.kicinski@...ronome.com> wrote:
>
> On Wed, 18 Sep 2019 17:37:44 -0400, Pooja Trivedi wrote:
> > Hi Jakub,
> >
> > I have explained one potential way for the race to happen in my
> > original message to the netdev mailing list here:
> > https://marc.info/?l=linux-netdev&m=156805120229554&w=2
> >
> > Here is the part out of there that's relevant to your question:
> >
> > -----------------------------------------
> >
> > One potential way for race condition to appear:
> >
> > When under tcp memory pressure, Thread 1 takes the following code path:
> > do_sendfile ---> ... ---> .... ---> tls_sw_sendpage --->
> > tls_sw_do_sendpage ---> tls_tx_records ---> tls_push_sg --->
> > do_tcp_sendpages ---> sk_stream_wait_memory ---> sk_wait_event
>
> Ugh, so do_tcp_sendpages() can also release the lock :/
>
> Since the problem occurs in tls_sw_do_sendpage() and
> tls_sw_do_sendmsg() as well, should we perhaps fix it at that level?

That won't do because tls_tx_records also gets called when completion
callbacks schedule delayed work. That was the code path that caused
the crash for my test. Cavium's nitrox crypto offload driver calling
tls_encrypt_done, which calls schedule_delayed_work. Delayed work that
was scheduled would then be processed by tx_work_handler.
Notice in my previous reply,
"Thread 2 code path:
tx_work_handler ---> tls_tx_records"


"Thread 2 code path:
tx_work_handler ---> tls_tx_records"