Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <92c2d051-e25f-7b3a-c811-dd1ce85f1b9b@linux.ibm.com>
Date:   Tue, 25 Jan 2022 13:35:23 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.4.0
Subject: Re: [RFC PATCH] uaccess: Add mechanism for key checked access to user
 memory
Content-Language: en-US
To:     Heiko Carstens <hca@linux.ibm.com>
Cc:     Christian Borntraeger <borntraeger@linux.ibm.com>,
        Vasily Gorbik <gor@linux.ibm.com>,
        Sven Schnelle <svens@linux.ibm.com>,
        Nico Boehr <nrb@linux.ibm.com>,
        Alexander Gordeev <agordeev@linux.ibm.com>,
        linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org
References: <YerCfDceDszqbdHU@osiris>
 <20220124103812.2340666-1-scgl@linux.ibm.com> <Ye7kuJ51QWFBGoJ4@osiris>
From:   Janis Schoetterl-Glausch <scgl@linux.ibm.com>
In-Reply-To: <Ye7kuJ51QWFBGoJ4@osiris>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 1/24/22 18:41, Heiko Carstens wrote:
> On Mon, Jan 24, 2022 at 11:38:12AM +0100, Janis Schoetterl-Glausch wrote:
>> KVM on s390 needs a mechanism to do accesses to guest memory
>> that honors storage key protection.
>> __copy_from/to_user_with_key is implemented by introducing
>> raw_copy_from/to_user_with_key.
>> Since the existing uaccess implementation on s390 makes use of move
>> instructions that support having an additional access key supplied,
>> we can implement raw_copy_from/to_user_with_key by enhancing the
>> existing implementation.
>>
>> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
>> ---
>>
>> This works for us and compiles on other architectures (tested x86).
>> The patch only implements __copy_from/to_user_with_key, since those
>> are the ones we actually need. On other architectures those functions
>> don't exists, but they aren't used either, so it's not a problem.
> 
> Adding an API where only underscored function names are to be used can be
> considered suboptimal.
> 
>> Should we also implement single and no underscore variants? Why?
>> Completeness?
> 
> Please make this _fully_ symmetrical to the existing copy_to/from_user()
> implementations, like I tried to say several times. Maybe I wasn't clear
> enough about this. Also the default implementation - that is if an
> architecture makes use of copy_to_user_key() without providing a
> raw_copy_from_user_key() implementation - should fallback to regular
> copy_to_user() semantics, like I tried to outline with the ifndef example
> of raw_copy_from_user_key() previously.

That does help. One thing I'm still confused about is the rational
for the default implementation.
Are you suggesting that copy_from/to_user be implemented in terms of
copy_from/to_user_with_key? I didn't think so, even tho you said something along
those lines, because I assumed you were referring to the architecture specific
implementations for copy_from/to_user, since we weren't talking about
common code changes back then and Christian's suggestion didn't feature it either.

When you say "fully symmetrical" do you mean all functions that wrap architecture
defined access to user space:
__copy_from_user_inatomic
__copy_from_user
__copy_to_user_inatomic
__copy_to_user
_copy_from_user
_copy_to_user
copy_from_user
copy_to_user
__copy_from_user_inatomic_nocache
copy_struct_from_user
copy_from_user_nofault
copy_to_user_nofault
strncpy_from_user_nofault
strnlen_user_nofault
> 
> Furthermore this should be splitted into two patches: one which adds the
> common code infrastructure, like described above; and a second patch which
> adds the actual s390 architecture backend/override.
> 
> The patches should contain a _detailed_ description why the first patch,
> aka API, should probably be in common code (staying in sync with code
> instrumentation, etc.); and of course it should contain enough information
> for people not familiar with s390's storage keys so they can figure out
> what this is about.
> 
> Hopefully we get some feedback and either this is acceptable for common
> code one way or the other, or we have to maintain this on our own, and get
> the additional maintenance cost for free.
> 
> Please make sure to add Al Viro, Kees Cook, Arnd Bergmann, and Andrew
> Morton to cc on your next version, so we hopefully come to a conclusion and
> can move on.

Thanks, will do.