Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220818110859.1918035-1-jens.wiklander@linaro.org>
In-Reply-To: <20220818110859.1918035-1-jens.wiklander@linaro.org>
From:   Linus Torvalds <torvalds@linuxfoundation.org>
Date:   Thu, 18 Aug 2022 09:38:14 -0700
Message-ID: <CAHk-=wiEqhzrMskE=7_q4E+X+zPyxm6xvuADhaBxh8gD5p57kA@mail.gmail.com>
Subject: Re: [PATCH v2] tee: add overflow check in register_shm_helper()
To:     Jens Wiklander <jens.wiklander@linaro.org>,
        linux-arch <linux-arch@vger.kernel.org>
Cc:     linux-kernel@vger.kernel.org, op-tee@lists.trustedfirmware.org,
        Sumit Garg <sumit.garg@linaro.org>, stable@vger.kernel.org,
        Nimish Mishra <neelam.nimish@gmail.com>,
        Anirban Chakraborty <ch.anirban00727@gmail.com>,
        Debdeep Mukhopadhyay <debdeep.mukhopadhyay@gmail.com>,
        Jerome Forissier <jerome.forissier@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Aug 18, 2022 at 4:09 AM Jens Wiklander
<jens.wiklander@linaro.org> wrote:
>
> Fix this by adding an overflow check when calculating the end of the
> memory range. Also add an explicit call to access_ok() in
> tee_shm_register_user_buf() to catch an invalid user space address
> early.

I applied the access_ok() part of this which was clearly missing.

The check_add_overflow() should be pointless with that.

And the "roundup() overflows" check should just check for a zero
result - if it is actually needed. Which I don't think it is on any
relevant platform (the TEE subsystem only works on arm and x86).

I do think it might be worth discussing whether
ALTERNATE_USER_ADDRESS_SPACE (and no-MMU) architectures should still
have access_ok() check that it doesn't actually wrap around in the
address space, so I've added linux-arch here.

That's m68k, PA-RISC, S390 and sparc.

In fact, I wonder if some or all of those might want to have the
TASK_SIZE limit anyway - they may have a separate user address space,
but several ones have some limits even then, and probably should have
access_ok() check them rather than depend on the hardware then giving
page fault.

For example, sparc32 has a user address space, but defines TASK_SIZE
to 0xF0000000. m68k has several different case. parisc also has an
actual limit.

And s390 uses

    #define TASK_SIZE_MAX         (-PAGE_SIZE)

which is a good value and leaves a guard page at the top.

So I think the "roundup overflows" would probably be best fixed by
just admitting that every architecture in practice has a TASK_SIZE_MAX
anyway, and we should just make access_ok() check it.

              Linus