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;
Subject: Re: [RFC PATCH 0/5] madvise MADV_DOEXEC
To:     "Eric W. Biederman" <ebiederm@xmission.com>,
        Anthony Yznaga <anthony.yznaga@oracle.com>
Cc:     linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        linux-mm@kvack.org, linux-arch@vger.kernel.org, mhocko@kernel.org,
        tglx@linutronix.de, mingo@redhat.com, bp@alien8.de, x86@kernel.org,
        hpa@zytor.com, viro@zeniv.linux.org.uk, akpm@linux-foundation.org,
        arnd@arndb.de, keescook@chromium.org, gerg@linux-m68k.org,
        ktkhai@virtuozzo.com, christian.brauner@ubuntu.com,
        peterz@infradead.org, esyr@redhat.com, jgg@ziepe.ca,
        christian@kellner.me, areber@redhat.com, cyphar@cyphar.com
References: <1595869887-23307-1-git-send-email-anthony.yznaga@oracle.com>
 <87pn8glwd2.fsf@x220.int.ebiederm.org>
From:   Steven Sistare <steven.sistare@oracle.com>
Organization: Oracle Corporation
Message-ID: <28125570-4129-bcba-099b-f90481cfbfe8@oracle.com>
Date:   Mon, 27 Jul 2020 14:00:17 -0400
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <87pn8glwd2.fsf@x220.int.ebiederm.org>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 7/27/2020 1:07 PM, ebiederm@xmission.com wrote:
> Anthony Yznaga <anthony.yznaga@oracle.com> writes:
> 
>> This patchset adds support for preserving an anonymous memory range across
>> exec(3) using a new madvise MADV_DOEXEC argument.  The primary benefit for
>> sharing memory in this manner, as opposed to re-attaching to a named shared
>> memory segment, is to ensure it is mapped at the same virtual address in
>> the new process as it was in the old one.  An intended use for this is to
>> preserve guest memory for guests using vfio while qemu exec's an updated
>> version of itself.  By ensuring the memory is preserved at a fixed address,
>> vfio mappings and their associated kernel data structures can remain valid.
>> In addition, for the qemu use case, qemu instances that back guest RAM with
>> anonymous memory can be updated.
> 
> What is wrong with using a file descriptor to a possibly deleted file
> and re-mmaping it?
> 
> There is already MAP_FIXED that allows you to ensure you have the same
> address.

MAP_FIXED blows away any existing mapping in that range, which is not the 
desired behavior.  We want to preserve the previously created mapping at
the same VA and co-exist with other mappings created by the new process.
There is no way to guarantee availability of a VA post-exec.

> I think all it would take would be a small protocol from one version
> to the next to say map file descriptor #N and address #A.  Something
> easily passed on the command line.
> 
> There is just enough complexity in exec currently that our
> implementation of exec is already teetering.  So if we could use
> existing mechanisms it would be good.
> 
> And I don't see why this version of sharing a mmap area would be
> particularly general.  I do imagine that being able to force a
> mmap area into a setuid executable would be a fun attack vector.

Any mmap(MAP_ANON) segment can be preserved.  That is very general, and is 
the case we need to support to upgrade legacy applications that are already
running (such as qemu) -- we cannot recode them before they are updated.

> Perhaps I missed this in my skim of these patches but I did not see
> anything that guarded this feature against an exec that changes an
> applications privileges.

The ELF opt-in flag must be set, so only selected executables will accept
incoming mappings.  The exec() code verifies that the preserved mappings 
do not overlap or become adjacent to text or stack, so it is not possible for
example for an attacker to cause stack underflow or overflow to access injected
content.  A gadget invoked by some other attack could access the preserved content.

- Steve

>> Patches 1 and 2 ensure that loading of ELF load segments does not silently
>> clobber existing VMAS, and remove assumptions that the stack is the only
>> VMA in the mm when the stack is set up.  Patch 1 re-introduces the use of
>> MAP_FIXED_NOREPLACE to load ELF binaries that addresses the previous issues
>> and could be considered on its own.
>>
>> Patches 3, 4, and 5 introduce the feature and an opt-in method for its use
>> using an ELF note.
>>
>> Anthony Yznaga (5):
>>   elf: reintroduce using MAP_FIXED_NOREPLACE for elf executable mappings
>>   mm: do not assume only the stack vma exists in setup_arg_pages()
>>   mm: introduce VM_EXEC_KEEP
>>   exec, elf: require opt-in for accepting preserved mem
>>   mm: introduce MADV_DOEXEC
>>
>>  arch/x86/Kconfig                       |   1 +
>>  fs/binfmt_elf.c                        | 196 +++++++++++++++++++++++++--------
>>  fs/exec.c                              |  33 +++++-
>>  include/linux/binfmts.h                |   7 +-
>>  include/linux/mm.h                     |   5 +
>>  include/uapi/asm-generic/mman-common.h |   3 +
>>  kernel/fork.c                          |   2 +-
>>  mm/madvise.c                           |  25 +++++
>>  mm/mmap.c                              |  47 ++++++++
>>  9 files changed, 266 insertions(+), 53 deletions(-)