Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Reply-To: crecklin@redhat.com
Subject: Re: [PATCH] security/keyring: avoid pagefaults in
 keyring_read_iterator
To:     David Howells <dhowells@redhat.com>
Cc:     Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>,
        James Morris <jmorris@namei.org>,
        "Serge E . Hallyn" <serge@hallyn.com>, keyrings@vger.kernel.org,
        linux-security-module@vger.kernel.org,
        linux-kernel@vger.kernel.org, Waiman Long <longman@redhat.com>
References: <20191018184030.8407-1-crecklin@redhat.com>
 <30309.1571667719@warthog.procyon.org.uk>
From:   Chris von Recklinghausen <crecklin@redhat.com>
Organization: Red Hat
Message-ID: <b8aa0f7c-0a90-efae-9fb7-aa85b19a0d9a@redhat.com>
Date:   Mon, 21 Oct 2019 11:46:15 -0400
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <30309.1571667719@warthog.procyon.org.uk>
Content-Language: en-US
Content-Type: text/plain; charset=WINDOWS-1252
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 10/21/2019 10:21 AM, David Howells wrote:
> Chris von Recklinghausen <crecklin@redhat.com> wrote:
>
>> The put_user call from keyring_read_iterator caused a page fault which
>> attempts to lock mm->mmap_sem and type->lock_class (key->sem) in the rev=
erse
>> order that keyring_read_iterator did, thus causing the circular locking
>> dependency.
>>
>> Remedy this by using access_ok and __put_user instead of put_user so we'=
ll
>> return an error instead of faulting in the page.
> I wonder if it's better to create a kernel buffer outside of the lock in
> keyctl_read_key().  Hmmm...  The reason I didn't want to do that is that
> keyrings have don't have limits on the size.  Maybe that's not actually a
> problem, since 1MiB would be able to hold a list of a quarter of a millio=
n
> keys.
>
> David
>

Hi David,

Thanks for the feedback.

I can try to prototype that, but regardless of where the kernel buffer
is allocated, the important part is causing the initial pagefault in the
read path outside the lock so __put_user won't fail due to a valid user
address but page backing the user address isn't in-core.

I'll start work on v2.

Thanks,

Chris